Big Structure is built on a foundation of reference structures, with domain structures capturing the domain at hand. These represent the target foundations for mapping schema and transforming data in the wild into an operable, canonical form. Any structure, even the most lightweight of lists and metadata, can contribute to and be mapped into this model, as this wall of structure shows:
Described below are some of these structures, in rough descending order of completeness and usefulness, for making data interoperable. Please note that any of these structures might be available as linked data.Reference Structures
In both semantics and artificial intelligence — and certainly in the realm of data interoperability — there is always the problem of symbol grounding. In the conceptual realm, symbol grounding means that when we use a term or phrase we are referring to the same thing. In the data value realm, symbol grounding means that when we refer to an object or a number, we are referring to the same measure.
UMBEL is the standard reference ontology used by Structured Dynamics. It contains 28,000 concepts (classes and relationships) derived from the Cyc knowledge base. The reference concepts of UMBEL are mapped to Wikipedia, schema.org (used in Google’s knowledge graph), DBpedia ontology classes, GeoNames and PROTON. Similar reference structures are used to ground the actual data values and attributes.
Other reference structures may be used, so long as they are rather complete in scope and coherent in their relationships. Logical consistency is a key requirement for grounding.Knowledge Bases
Knowledge bases combine schema with data in a logical manner; well-constructed ones support computations, inference and reasoning. To date, the two primary knowledge bases that we use are Wikipedia and Cyc. However, many specific domain knowledge bases also exist.
Knowledge bases are important sources for symbol grounding. It addition, because of their computability, they may be used with artificial intelligence methods to both extend the knowledge base and to refine the feature estimates used in the AI algorithms.Domain Ontologies
Domain ontologies, constructed as graphs, are the principal working structures in data interoperability. Though best practices recommend they be grounded in the reference structures, the domain structures are the ones that specifically capture the concepts and data attributes of the target information domain. More effort should be focused at this level in the wall of structure than any other.
Domain structures provide unique benefits in discovery, flexible access, and information integration due to their inherent connectedness. Further, these domain structures can be layered on top of existing information assets, which means they are an enhancement and not a displacement for prior investments. And, these domain structures may be matured incrementally, which means their development is cost-effective.Mappings
Data and schema in the wild need to be mapped and transformed into these canonical structures. What is known as data wrangling is an aspect of these mappings and transformations. Mappings thus become the glue that ties native data to interoperable forms.
Mapping is the critical bridging function in data interoperability. It requires tools and background intelligence to suggest possible correspondences; how well this is done is a key to making the semi-automatic mapping process as efficient as possible. Mapping structures are the result of the final correspondences. Mapping effort is a function of the scope of Big Structure, not the volume of Big Data.Existing Structures
A broad variety of structures occur in the wild — from database schema and taxonomies to dictionaries and lists — that need to be represented in a common form and then mapped in order to support interoperability. The common representation used by Structured Dynamics is the RDF data model.Structure Scripts
Scripting and tooling are essential to help create Big Structure efficiently.Editor’s Note: We are pleased to share with you in advance some of the text from Structured Dynamics’ new Web site.
Structured Dynamics today released version 1.10 of its open source UMBEL (Upper Mapping and Binding Exchange Layer) reference concept ontology. This new release is in preparation for a series of subsequent releases planned over the next few months as additional consistency and functionality is brought to the system.
This is the first UMBEL version to be created and loaded from scratch using a new Clojure scripting framework. Fred Giasson describes these scripts and highlights some of the new functionality in a separate blog post.
Besides this new processing framework, here are the key changes at a high level made in this new version:
This new UMBEL v. 1.10 is now updated and consistent with OpenCyc (version 4.0, dated October 12, 2012), GeoNames (version 3.1, dated October 29, 2012), schema.org (version 1.9, dated August 19, 2014) and the DBpedia ontology (version 3.9, retrieved August 18, 2014). The resulting mappings now are:
These changes have also resulted in some improvements to the umbel.org Web site and its Web services, as more fully described by Fred. The updated ontologies and mappings may be found at the UMBEL GitHub site.What’s In the Pipeline
As noted, this version update is in preparation for some pending activities, which are now moving toward completion. Subsequent releases (perhaps not in the order shown) will include:
Look here and on the UMBEL mailing list for these future announcements.
In the earlier installments of this article series we first described how to estimate the value of connections amongst Big Data datasets, premised on the network effect. We then went into detail in the second part about the Viking algorithm (VKG) derived to capture the Value of Knowledge Graphs.
In this concluding part of the series we summarize the use and implications of the Viking algorithm on your Big Data planning. We do so by offering ten guidelines for how including Big Structure may be leveraged in the context of a knowledge graph. We then conclude the series with some caveats on the interpretation of these results and a discussion of possible future directions.Ten Guidelines for Big Structure
As you think through your Big Data initiatives, we recommend you keep these ten guidelines relating to data structure in mind:
Of course, mere connections for structure’s sake is silly. It is important that the structure added and connections made are correct, consistent and coherent. Even then, not all types of connections are created equal, with typing the most important, annotations the least.
The good thing is that Big Structure can be added as a slight increase over standard data wrangling efforts, and with much greater impact than standard wrangling. Further, the structures themselves, preferably guided by domain ontologies, are a means of testing these factors for subsequent structure additions. Not only does adding structure get easier with a foundation of existing structure, but it increases the value of the information by orders of magnitude.Not the Last Word
Roughly twenty years ago Metcalfe’s law triggered a gold rush in trying to achieve network effects at Internet scale. Though the algorithm proved too optimistic at larger scales, the idea of the benefit from connections was firmly established. Ten years ago it was clear that some form of diminishing returns needed to be applied to connections at scale. Zipf’s law was not a bad guess, though we have subsequently learned that more graph-centric measures are more appropriate and accurate for estimating value. Now, the Viking algorithm has emerged as the best estimator of the value of connections within Big Data.
I suspect we will see further improvements to the Viking algorithm as: 1) we come to better understand graph structures (including the effects of clusters and cliques); and 2) we learn to distinguish the different value of different types of connections . We can already see that typing and categorization have better structural effects above annotations. We further can see that the correctness of asserted “facts” is a key to realizing the multiplier benefits of connections and structures. Thus, we should see improved means for screening and testing assertions for their accuracy at scale.
At this stage, what the Viking algorithm gives us is a defensible means for assessing the value of adding structure (through connections) to our datasets. We see these multiplier effects to be huge, and to compound to even still further benefits with scale. We also see that the most developed forms of structure — namely, ontologies — bring still further benefits in inference and testable coherence. All Big Structure efforts should be aiming to express all of the structural insights for the organization and its datasets into these ontological forms .
While our current proxy for value — namely, asserted “facts” — is useful, it would also be helpful to be able to translate these “fact” assertions into a monetary value. As we move down this path we will discover, again, that not all “facts” are created equal, and some have more monetary value than others. Transitioning our estimates of value to a monetary basis will help set parameters for the cost-benefit analysis of data collection and structurizing that is the ultimate basis for planning Big Data and Big Structure initiatives.
In the end, many things need to be analyzed to understand the impacts of each connection and structure metric on the value of the resulting graph. But, what today’s current understanding of the network effect and the Viking algorithm brings us is a better means to understand and quantify the benefits of connected information. By any measure, these value benefits are multiples of what we see for unconnected data, the multiples of which grow massively with the scale of the data and their connections.
Big Structure is fertile ground for bringing in the sheaves. Let the harvesting begin. Though not further discussed here, the ontologies also provide the means for tagging (providing structure) to unstructured documents, which also brings the multiplier benefits from structure. On the retrieval side, such structure also aids faceting and filtered “slicing and dicing” of underlying datasets, thereby improving retrieval efficacy.  As one of the first approaches to capture these nuances, see Mischa Dohler, Thomas Watteyne, Fabrice Valois and Jia-Liang Lu, 2008. Kumar’s, Zipf’s and Other Laws: How to Structure a Large-Scale Wireless Network?, published in Annales des Telecommunications – Annals of Telecommunications 63, 5-6 pp. 239-251. See http://hal.archives-ouvertes.fr/docs/00/40/58/67/PDF/Large_Scale_Networks_journal_FINAL.pdf.
Yesterday in the first part of this series we raised the important question of how to value connections made between data. At the Big Data scales represented, we prepared a Basic Facts case of up to 2 billion assertions. We are using asserted “facts” as our value proxy. We’ll talk more about value and caveats in the third part of our series, tomorrow.
We saw that early estimates of network effects, such as Metcalfe’s law, overestimate value at scale. We looked at Zipf’s law as a means to capture the diminishing value of connections given the distance between facts. In today’s article we will focus on these factors of interaction and potential value in the specific context of knowledge graphs. Knowledge graphs are Big Structure representations that capture the schematics, concepts and measures in any given knowledge domain (that is, any domain of human activity).
Since I first tried to address the value of knowledge networks some five years ago , I have been disturbed about a couple of things . First, I felt that the exponential or geometric bases for estimating the value of information connections were not correct, both because they fail at scale and they don’t discriminate that some connections work and are more important, while others are trivial or don’t work. Capturing this law of diminishing value in a context that makes sense for knowledge bases was, I felt, the key to answering the value riddle.
I believe we have now, in this series, provided a compelling basis for solving that riddle, which also points the way to further improvements. This assertion is an exciting statement, in that we now may have a quantitative basis in hand for determining where and how to spend our monies for Big Data and Big Structure initiatives. Such quantitative tools are a huge boon to bring analytic rigor to the data collection and integration challenge.Adding connections (“Big Structure“) to Big Data can increase the value of enterprise information from one to three orders of magnitude; the value also scales linearly with added structure (attributes).
This article shows that adding connections (“Big Structure“) at Big Data scales can increase the value of enterprise information from one (ten) to three (thousands) orders of magnitude. The magnitude of the value scales linearly with each added structure (attribute). These value multipliers from adding Big Structure are a tremendously cost-effective addition to standard data wrangling efforts.The Value of Knowledge Graphs (VKG) Formulation
The recognition of the need for a law of diminishing returns to reflect the distance between facts or assertions is a central argument in the Briscoe-Odlyzko-Tilly formulation (see  directly, and the prior Part I discussion). Not all information is connected, and not all connected information is of equivalent worth. The implied question in these statements, however, is how to capture those differences?
The B-O-T (or sometimes, O-T) formulation does not choose a bad starting proxy for this diminishment law. Zipf’s law reflects many observed distributions in human objects, roughly equivalant to power law, Pareto (“80:20″) distributions or n log (n) diminishing returns with long-tail characteristics. Examples include Internet distributions (such as popularity of Web sites or search terms), human language distributions, income rankings, population distributions, etc. There is no question that Zipf’s law distributions are common and frequent.
The only problem with picking the Zipf’s law basis, however, is that there was absolutely no evidence that such occurred for information networks or knowledge graphs. Zipf’s law distributions tend to be statements across single types for a single attribute distribution. Graphs, we can safely say, are anything but this distribution. Connections and multiple types are the rule, not the exception.
So, maybe the B-O-T formulation was correct, and maybe it was not. There was no empirical evidence to support this assertion for knowledge graphs. And, there did not appear to be a compelling logic argument for relating Zipf’s law to graphs other than they are artifacts of human endeavor.
My discomfort in adopting this arbitrary B-O-T basis, even though solidly embedded in human experience, caused me to seek alternative ideas and explanations, but also ones that fulfilled the key structural insights of diminishing returns and non-equivalent assertions that were the focal points of B-O-T, all within a graph context.The Starting Basis
The breakthough occurred when I discovered an obscure, un-cited paper by Yaakov Stein . Stein, a network and signals processing researcher of the first rank , wrote his paper as a means to understand and quantify his experience of joining LinkedIn and expanding his network. He began without an account and documented his experience as he joined and expanded his network of contacts on LinkedIn. He charted direct links, and then meticulously looked at and recorded secondary and tertiary links.
His formulation recognized that the value to an individual user equaled raising the access to the entire network (1) for that user plus the diminishing benefit represented by the participating graph’s other participants as measured by average degree of separation (d). d is an inherent measure of the graph type.
Though his context was a social network, the basic observation obtains: relations diminish by distance within a graph, with average link distance (directly related to degree of separation) being the key relevance metric. Connected “facts” or “friends” is essentially the same thing. It is all about what is shared amongst graph nodes.
The usefulness of this approach is that it grounds the multiplier effect in an inherent characteristic of the source graph, its average degree of separation . Like Zipf’s law, the degree of separation is a distance measure, but one grounded specifically in graphs. Here is the Stein formulation:
A graph with a degree of separation of 4, then, would exhibit a network-wide power factor of 5/4 (4/4 plus 1/4).The Viking Algorithm
As applied to knowledge graphs, however, this formulation still has two problems. The first minor one is that the degree of separation parameter should be D (the average across structure) rather than d. The second substantive one is that a correction factor needs to be included that accounts for the probability that an assertion may be false. This factor, F, is 1 – the measured error rate.
The resulting algorithm we term the Value of Knowledge Graph formulation, or the VKG (Viking) algorithm. It is expressed like this:
F is meant to be analogous to F-measure, the combined precision and recall statistic for information retrieval and NLP tasks. F in the case of the Viking algorithm is also meant to be a combined statistic that represents the “accuracy” (verifiable truthfulness) of statements asserted in the graph. F is essentially an estimated value for the residual falsity for the average statement in a graph, after removal of all assertions that do not meet existing coherency, consistency or completeness tests. F is determined by sampling statements across the graph and manually testing for truthfulness (or in a logical sense, validity given the existing statements in the graph). An F of 1 signifies complete truthfulness (accuracy); an F of 0 represents complete falsity .Viking in Relation to Other Network Estimators
Now, with this explanation of basis, we can again look at the value of the Viking (VKG) algorithm in comparison to those discussed in the first part of this series. Again on a logarithmic scale, here are those results:
Excluding the exponential and geometric multipliers (namely, the “laws” of Metcalfe and Reed) in the top two curves, this shows the Viking (VKG) algorithm to have higher value than the B-O-T (O-T) algorithm, both of which are considerably higher than the Basic Facts. However, because the Figure 1 above has a logarithmic scale, these differences are harder to discern.Viking Benefits Over the Basic Facts
Now corrected with our assumed F factor, we can begin to tease out the value benefits of connecting “facts” versus the unconnected Basic Facts. As with any logarithmic function, we see that the value benefits from connections increase in a growing manner at larger scales. For example, as Figure 2 shows below, at a level of 1000 records, the benefits from connections are 7x greater than unconnected data. By the time the scale grows to 1 million or 500 million records, the value benefits of connections grows to 44x to 215x, respectively:
Benefits from connections increase as a power function at increasing scales.Setting the VKG Factor D
But the potential value of connectedness is also a function of the general degree of information separation for the given domain. We are still in the early phases of gathering statistics for such things, but the table below summarizes what is known about the “standard” level of connectivity in various domains and applications. Note, in general, most any knowledge graph would have a D factor ranging from 2 to 8:Category Degrees of Separation (D) Notes Food webs ~ 2  Genetic differences ~ 3  LinkedIn ~ 3  Twitter 3.435 – 4.67  Facebook 3.74  Potential research collaborators ~ 4  UMBEL ~ 5.2  Social networks (general) ~ 6 Mobile ad hoc networks ~ 7  Small-world networks (max) ~ 8  Table 1. Degrees of Separation for Various Knowledge Networks
More tightly linked, cohesive domains tend to have the lower degrees of separation. It is also interesting to note that some social networks, like Twitter and Facebook, are also able to lower degrees of separation (in comparison to their nominal “social network” benchmark) by virtue of the nature of their service.
As experience is gained and with more research, I expect more estimates and more refined ones. Depending on the nature of the domain at hand, it should then be possible to pick the closest analog to use in the Viking valuation algorithm. Nonetheless, we already have a range and respective values to provide meaningful value estimates today.
Using the values in Table 1, we are thus able to plot the effects (again, log scale) of these various degrees of separation in terms of the “fact” assertions that can be made for our Big Data test dataset:
At the nominal Big Data scales of 100,000 and 1,000,000 records, the value of data connections in comparison to the unconnected Basic Facts case shows these following value improvement multipliers:Domain 100,000 Records 1,000,000 Records Food webs 203x 611x Genetic differences 38x 84x Twitter 23x 46x Facebook 17x 33x Potential research collaborators 14x 26x UMBEL 8x 12x Social networks (general) 5x 8x Mobile ad hoc networks 3x 5x Table 2. Multiplier (X) Improvements by Domain from Connections Over the Basic Facts
Of course, our “Big Data Example” from Part I was silent about the exact nature of its knowledge graph. Based on empircial experience to date, the benefits from connecting data that was previously unconnected should fall somewhere within the limits of Table 2. Even at rather low scales and more loosely-connected domains, the value improvements in making connections with data is many-fold. At larger scales for tighter networks, the multipliers can become astounding.Adding Structure to the Underlying Data
Another implication that the Viking algorithm allows us to test is the comparative benefit from adding structure to our datasets. Actually, “adding structure” is not strictly correct; it is “structurizing” the data via characterizations, attributes and categorizations. Of course, not all structure is created equal. Assigning or classifying our records into types, for example, applies to all records across the datasets and provides powerful cross-record linkages. Adding annotations or metadata to single records provides much lower benefits.
When we add structure across datasets the value improvements are a linear percent, as this figure shows:
For our Big Data example, each across-dataset structure characterization adds about 25% to 30% value per structure. Adding four structural characterizations, for example, more than doubles the “facts” assertion value (~ 140%) to the datasets.Preview of Last Part
The last part forthcoming tomorrow will summarize the implications from the Viking algorithm on the role and importance of Big Structure to your organization’s Big Data efforts. Some caveats and future directions will conclude the series. The two articles written at that time were, M.K. Bergman, 2009. Structure the World, in AI3:::Adaptive Information blog, August 3, 2009, and M.K. Bergman, 2009, The Law of Linked Data, AI3:::Adaptive Information blog, October 11, 2009.  See also the then-current state of analysis by Eric Hellman, 2009. Normal and Inverse Network Effects for Linked Data, published in his blog, October 15, 2009; see http://go-to-hellman.blogspot.com/2009/10/normal-and-inverse-network-effects-for.html.  Bob Briscoe, Andrew Odlyzko, and Benjamin Tilly, 2006. “Metcalfe’s Law is Wrong,” in IEEE Spectrum, July 2006. A copy may be viewed at http://www.cse.unr.edu/~yuksem/teaching/nae/reading/2006-briscoe-metcalfes.pdf. Odlyzko, and Tilly had published an earlier version, (sometimes the approach is shown as O-T in addition to B-O-T), and the basic form of the algorithm appears in a single Odlyzko paper.  Yaakov (J) Stein, 2009. The Value of Being Linked In, on his personal Web site, April 2009; see http://www.dspcsp.com/pubs/linkedin.pdf. Note that his empirical tests suggested a degree of separation for LinkedIn of 3.  The average degree of separation is simply the graph’s average path distance – 1. For an explanation of average path distance, see .  F is a summed average value across all assertions within a knowledge graph. In information retrieval, F-measures are now being achieved that exceed 0.90 (90%). For the cases used herein, F is estimated at 0.85. Again, this parameter is measured after all standard coherency, consistency, and completeness tests are applied to the ontology. These tests routinely remove many false assertions and establish the basic integrity of the graph. This acceptance threshold is itself constantly improving as experience is gained with basic graph integrity tests. In other words, tomorrow’s thresholds will be higher than today’s.  Richard J. Williams, Eric L. Berlow, Jennifer A. Dunne, Albert-László Barabási, and Neo D. Martinez, 2002. Two Degrees of Separation in Complex Food Webs, in Proceedings of the National Academy of Sciences, 99 (20):12913-12916, September 16, 2002, doi:10.1073/pnas.192448799; see http://www.pnas.org/content/99/20/12913.full.  See http://blog.23andme.com/ancestry/three-genetic-degrees-of-separation/.  Reza Bakhshandeh, Mehdi Samadi, Zohreh Azimifar and Jonathan Schaeffer, 2011. Degrees of Separation in Social Networks, in Proceedings, The Fourth International Symposium on Combinatorial Search (SoCS-2011), 6 pp.; see http://www.aaai.org/ocs/index.php/SOCS/SOCS11/paper/viewFile/4031/4352; and Haewoon Kwak, Changhyun Lee, Hosung Park, and Sue Moon, 2010. What is Twitter, a Social Network or a News Media?, in Proceedings of the 19th International Conference on World Wide Web, April 26–30, 2010, Raleigh, North Carolina, pp. 591-600, ACM; see http://snap.stanford.edu/class/cs224w-readings/kwak10twitter.pdf.  Lars Backstrom et al., 2012. Four Degrees of Separation, Archiv.org, January 6, 2012; see http://arxiv.org/pdf/1111.4570.pdf.  Paweena Chaiwanarom, Ryutaro Ichise, and Chidchano Lursinsap, 2010. Finding Potential Research Collaborators in Four Degrees of Separation, pp. 399-410, in Longbing Cao, Jiang Zhong, and Yong Feng, eds., Advanced Data Mining and Applications, Springer Berlin Heidelberg, http://dx.doi.org/10.1007/978-3-642-17313-4_39.  See http://fgiasson.com/blog/index.php/2014/08/11/graph-analysis-of-a-big-structure-umbel/#Average-Path-Length-Distribution.  Maria Papadopouli and Henning Schulzrinne, 2000. Seven Degrees of Separation in Mobile ad hoc Networks, presented at Global Telecommunications Conference, 2000 (GLOBECOM’00), IEEE. Vol. 3; see http://www.huaxiaspace.net/academic/classes/wi02/cse294/20020222globecom2000.pdf.  Paolo Pin, 2006. Eight Degrees of Separation, in Nota di Lavoro, Fondazione Eni Enrico Mattei, No. 78.2006 see http://www.econstor.eu/bitstream/10419/74249/1/NDL2006-078.pdf.
The hackneyed phrase of “connect the dots” reflects our basic intuition that there is value in making connections amongst relevant data. But, what is this value? How might we quantify it? This topic, and a method and guidelines for doing so, are the subject of this article and the second and third parts to follow.
The reason it is important to quantify the value of connected information is that such an estimate helps to define what effort or cost we can justify in order to derive those connections. In Big Data, for example, we already know that 50% to 80% of the costs in assembling relevant datasets is due to data wrangling — the effort to extract, transform and clean the input data . No where, however, do we know what it is worth to go to the next step of working to connect those data.
Quantifying this understanding will thus also help determine what the value is in developing Big Structure, the approach we have been most recently discussing for how to organize and connect Big Data. Big Structure sets the schematic and data relationships for how data from disparate sources can be connected together.
About five years ago I wrote my first articles on how we might approach the quantification of these information connections . That first, cursory look was useful for bounding the problem, but no firm conclusions as to how to specifically quantify this value were proposed. Like other graphs or networks, the usefulness of the ‘network effect‘ to bound the question was clear. It has taken further research and experiences with actual linked datasets to point to how to resolve this quantification challenge.Foundations in the Network Effect
The network effect was first realized in the early days of telephone networks, where the value of the system increased as a function of more users . We have also long recognized a similar effect in connecting information together and the breaking down of information or ‘data silos‘. As the following diagram shows, unconnected data nodes or silos look like random particles caught in the chaos of Brownian motion:
As initial connections get made, bits of structure begin to emerge. But, as connections are proliferated — analogous to the network effects of connected networks — coherence and more structure emerge.
This emergence of structure is particularly evident in physical networks, such as the growth of this hypothetical telecommunications network:
This diagram, modified from Wikipedia to be a horizontal image, shows how two telephones can make only one connection, five can make 10 connections, and twelve can make 66 connections, etc. It is this very multiplier effect that has led to most of the thinking of how to quantify the network effect.
We can see an interesting parallel between telecommunications networks and knowledge graphs. In the telecommunication network, the addition of a new user (node) by definition brings with it connections. This is what is shown in Figure 2. But in information silos, the information is already there (nodes, or the left-side of Figure 1); what is missing are the connections (the right-side of Figure 1). By explicitly adding connections we can also create network effects, as others have noted .
However, once we understand these parallels, we must also recognize the differences. To properly estimate the network effects of knowledge graphs, we must be explicit about the similarities and differences with other (physical) networks .Objectives for a Knowledge Graph Formulation
Since our objective is to quantify the “value” of a knowledge graph, we must first ask what is the basis of this value. In the best of all worlds, we would know the monetary worth of information, so we could justify what to spend in order to leverage it, which of course varies wildly across bases and sources. But we don’t. We do know, however, that a knowledge graph and the information it connects to constitutes a knowledge base. In the context of a knowledge base, the measure of value is the number of “correct” facts it contains. Therefore, we will use the number of connections in the graph (equivalent to the number of triple statements) as a proxy for value, representing the asserted “facts” of the graph.
We will also seek measures of graph distance and connectedness to capture the network-like qualities of the knowledge base. The characteristics of the graph itself should be the input base upon which to estimate value.Alternative Estimates of the Network Effect
The earliest effort to estimate the value of physical networks was Sarnoff’s law, developed by David Sarnoff, for many years the leader of the Radio Corporation of America (RCA). He posited that the value of a broadcast network was directly proportional to its number of viewers (n). However, the problem with this formulation is that a broadcast network is only one way, from broadcaster to user. What of networks where there is interaction or two-way linkages? The benefits of such networks must surely be more than linear.
Once we get into interaction effects we get into multipliers. And the proper nature of those multipliers must come from the nature and extent of those interactions, as well as perhaps the nature of the network itself. I discuss below some of the more prominent candidates that have been put forward for estimating the network effect, or the value of networks.Metcalfe’s Law
Metcalfe’s law was the first direct derivation from the telecommunications model. Robert Metcalfe formulated it about 1980 in relation to Ethernet and fax machines; the “law” was then named for Metcalfe and popularized by George Gilder in 1993. The actual algorithm proposed by Metcalfe calculates the number of unique connections in a network with n nodes to be n(n − 1)/2, which is proportional to n2. This makes Metcalfe’s law a quadratic growth equation.
The law is generally simplified  to state that the value of a telecommunications network is proportional to the square of the number of users of the system (n²):
Gilder’s popularization and the early growth of the Internet made the estimation of the benefits of network effects a very timely topic. For example, as a value measure, the network effect could be used to estimate the benefits for larger and larger numbers of users. Some have even blamed Metcalfe’s law for contributing to the creation (and then bursting) of the “dot-com bubble” of the late 1990s .
Metcalfe’s law clearly showed that interaction effects between nodes could generate multipliers that scaled rapidly with increasing numbers of nodes (users).Reed’s Law
From a different perspective and with a different take, David Reed came up with a multiplier formulation that is the largest presented — anywhere. Reed’s context is social groups, and from that perspective he can envision arbitrary sized groups forming amongst any and all participants (nodes). Because of this theoretical, global scope, justified through examples such as eBay and chat rooms, Reed specifically defined group-forming networks (GFNs), as the applicable scope . The simplified formulation for Reed is:
In scope and context, Reed does not apply to knowledge graphs, and even in the areas of social groups, most researchers find the exponential implications of Reed’s law unsupportable . The next group, for example, offers direct criticism.Briscoe – Odlyzko – Tilly Formulation
Under the provocative title, “Metcalfe’s Law is Wrong,” Briscoe, Odlyzko, and Tilly challenged both the Metcalfe and Reed approaches in 2006 . Using the proxy of Internet valuation, the authors were able to show how impractical the implications of either approach were at scale. Like the bet of rice (or wheat) doubling each of the 64 squares on a chessboard bankrupting the kingdom, the exponential implications of these two “laws” can be seen to (eventually) violate common sense.
The fundamental fallacy associated with both the Metcalfe and Reed approaches is that all potential links are of equal value . But no where in the real world do we see this to be true. There must be some law of diminishing returns that must be applied to slow the unsustainable rates of exponential or (to a lesser extent) quadratic growth.
After much hand waving, the authors chose Zipf’s law as their basis for this diminishing return. The increasing “decay rate” with distance is a common distribution pattern for real-world datasets, which Zipf’s law specifically addresses, always showing power law distributions with long tails. To approximate this distribution they offered the simple n log (n) formulation of Zipf’s law .
This is a reasonable approximation, but one that is never related directly to the nature of graphs or networks. That is the source of the next layer of refinements.VKG Formulation
I will discuss this algorithm, our recommended formulation, in Part II of this series.A Big Data Example
In order to discuss further the question of value arising from network effects, we need a case study example. We also need to define “value”, which in this case study example, as also noted above, means the number of “facts” or assertions in our database . This we call the Basic Facts (“assertions”) column.
For our basic “facts”, we consider a data series that is doubling in size for each step, eventually reaching a half billion records. Each record has four attributes or characterizations, leading to a total of more than 2 billion “facts” in the database. These are the first two columns in this table:Records Basic Facts (“assertions”) Big Data Example 1 4 4 2 8 8 4 16 16 8 32 32 16 64 64 32 128 128 64 256 256 128 512 512 256 1,024 1,024 512 2,048 2,048 1,024 4,096 4,100 2,048 8,192 8,200 4,096 16,384 16,400 8,192 32,768 32,800 16,384 65,536 65,600 32,768 131,072 131,200 65,536 262,144 262,404 131,072 524,288 524,812 262,144 1,048,576 1,049,624 524,288 2,097,152 2,099,248 1,048,576 4,194,304 4,198,496 2,097,152 8,388,608 8,396,996 4,194,304 16,777,216 16,793,992 8,388,608 33,554,432 33,587,984 16,777,216 67,108,864 67,175,972 33,554,432 134,217,728 134,351,944 67,108,864 268,435,456 268,703,888 134,217,728 536,870,912 537,407,780 268,435,456 1,073,741,824 1,074,815,564 536,870,912 2,147,483,648 2,149,631,128 Table 1. Basic Facts Connections with a Big Data Example
At this point, we have no connections between records. Each record has four attributes each, in isolation. This basis is akin to the unconnected dots on the left side of Figure 1 above.
For our Big Data Example, we will posit a record matching procedure as our first task for a new Big Data initiative. The assumption is that across all records, one-in-10000 matches another record. This results in the number of assertions (“facts”) shown in the third column in the table above. The posited Big Data initiative results in a 0.10% increase in “facts”, irrespective of record scale, once the matching threshold is reached. This result is not terribly impressive, but is perhaps not too unrelated from a first foray into a Big Data project.
Note that the following charts and analyses (including in the next part tomorrow) use as their “Basic Facts” the number of “assertions”, or the middle column in the table above. Though Big Data may represent an initial 0.10% improvement over this, that is immaterial to what our Big Structure viewpoints will provide. So, our “Basic Facts” will be unconnected records.Applying Network Effects to the Basic Facts
We can now apply our various network effect estimators to this base case. And, because of the fast-compounding nature of both the Reed and Metcalfe approaches, we need to plot this out on logarithmic scale  (click to enlarge):
On a logarithmic scale, the O-T (Briscoe-Odlyzko-Tilly) and VKG formulations appear only marginally better than the Basic Facts base case, but that is only due to the swamping effects of the unrealistic growth multipliers. We’ll get into this more tomorrow.Preview of Next Part
The next part forthcoming tomorrow will use this foundation to describe the VKG algorithm, and some of implications of its characteristics, all in the context of knowledge networks or graphs. “Data scientists, according to interviews and expert estimates, spend from 50 percent to 80 percent of their time mired in this more mundane labor of collecting and preparing unruly digital data, before it can be explored for useful nuggets,” is a quote from Steve Lohr, 2014, “For Big-Data Scientists, ‘Janitor Work’ Is Key Hurdle to Insights,” August 17, 2014, New York Times, see http://www.nytimes.com/2014/08/18/technology/for-big-data-scientists-hurdle-to-insights-is-janitor-work.html. Also, as another example of the common 80% estimate for data preparation costs, see http://radar.oreilly.com/2013/09/data-analysis-just-one-component-of-the-data-science-workflow.html.  These two articles were, M.K. Bergman, 2009. Structure the World, in AI3:::Adaptive Information blog, August 3, 2009, and M.K. Bergman, 2009, The Law of Linked Data, AI3:::Adaptive Information blog, October 11, 2009. The same concerns I had at that time in the current state of analysis was captured by Eric Hellman, 2009. Normal and Inverse Network Effects for Linked Data, published in his blog, October 15, 2009; see http://go-to-hellman.blogspot.com/2009/10/normal-and-inverse-network-effects-for.html.  These network effect benefits were reportedly a major driver of Theodore Newton Vail‘s efforts to consolidate the thousands of initial telephone networks in the United States under the banner of the American Telephone & Telegraph (Ma Bell) company.  See, for example, James Hendler and Jennifer Golbeck, 2008. Metcalfe’s Law, Web 2.0, and the Semantic Web, in Web Semantics: Science, Services and Agents on the World Wide Web 6(1): 14-20; see http://www.cs.umd.edu/~golbeck/downloads/Web20-SW-JWS-webVersion.pdf.  Babak Hodjat and Adam Cheyer, 2003. Evolution of the Laws that Deal with the Utilization of Information Networks, in Masoud Nikravesh, Lotfi A. Zadeh and Janusz Kacprzyk, eds., Studies in Fuzziness and Soft Computing, Vol 164/2005, pp. 427-438, Springer, Berlin. See http://www.adam.cheyer.com/papers/KnowledgeNetworks_Formatted.pdf.  For a well-connected network, every node (n) connects to every other node (n-1), which gives us n*(n-1) or (n2 – n). Working this out, two nodes have two connections (2*2 – 2), three nodes have six connections (3*3 – 3) and the expression converges on the square of ‘n’ for larger values of ‘n’, e.g., (100*100 – 100) is 99% of (100*100). This convergence at larger number is the basis for the exponential simplification, 2n. Most of the other ‘laws’ stated herein are simplifications in a similar manner.  See, for example, Sara F. Peralta, 2011. Moore’s Law, Metcalfe’s Law, and the Dot Com Bubble, November 27, 2011, see https://storify.com/sarafperalta/moore-s-law-metcalfe-s-law-bubble. Also see .  David P. Reed, 1999. That Sneaky Exponential—Beyond Metcalfe’s Law to the Power of Community Building, August 27, 1999; online at http://www.reed.com/dpr/locus/gfn/reedslaw.html. For original version, see http://contextmag.com/archives/199903/digitalstrategyreedslaw.asp. Like Metcalfe, at smaller numbers the actual formula is 2n -n – 1, which rapidly converges to 2n.  However, one group has published an alternative formulation consistent with the Reed approach; see Kalevi Kilkki, and Matti Kalervo, 2004. KK-law for Group Forming Services, in XVth International Symposium on Services and Local Access, Edinburgh, March 2004. See http://kotisivukone.fi/files/50ajatelmaa.ajatukset.fi/tiedostot/Others/kilkki_kk-law.pdf.  Bob Briscoe, Andrew Odlyzko, and Benjamin Tilly, 2006. “Metcalfe’s Law is Wrong,” in IEEE Spectrum, July 2006. A copy may be viewed at http://www.cse.unr.edu/~yuksem/teaching/nae/reading/2006-briscoe-metcalfes.pdf. Odlyzko, and Tilly had published an earlier version, (sometimes the approach is shown as O-T in addition to B-O-T), and the basic form of the algorithm appears in a single Odlyzko paper.  See, for example, https://www.princeton.edu/~achaney/tmve/wiki100k/docs/Zipf_s_law.html.  In specific terms, each “fact” in our knowledge base is an assertion, represented as an RDF triple statement. Because some of these assertions may not, in fact, be true, the use of “fact” does not imply universal truthfulness. Rather, an assertion that passes current tests for logic, coherency, consistency or completeness is what is retained, even though its truthfulness is not certain. Therefore, separate adjustment factors (parameters) need to be applied to address accuracy tests.  We adjusted the scale further to reduce the exponential absurdity of the Reed approach by manually shifting the scale downward. As a result, the Reed approach exits the chart rather quickly, heading straight up.
In the first parts of this series we introduced the idea of Big Structure, and the fact that it resides at the nexus of the semantic Web, artificial intelligence, natural language processing, knowledge bases, and Big Data. In this article, we look specifically at the work that Big Structure promotes in data interoperability as a way to clarify what the roles these various aspects play.
By its nature, data integration (the first step in data interoperability) means that data is being combined across two or more datasets. Such integration surfaces all of the myriad aspects of semantic heterogeneities, exactly the kinds of issues that the semantic Web and semantic technologies were designed to address. But resolving semantic differences can not be fulfilled by semantic technologies alone. While semantics can address the basis of differences in meaning and context, resolution of those differences or deciding between differing interpretations (that is, ambiguity) also requires many of the tools of artificial intelligence or natural language processing (NLP).
By decomposing this space into its various sources of semantic heterogeneities — as well as the work required in order to provide for such functions as search, disambiguation, mapping and transformations — we can begin to understand how all of these components can work together in order to help achieve data interoperability. This understanding, in turn, is essential to understand the stack and software architecture — and its accompanying information architecture — in order to best achieve these interoperability objectives.
So, this current article lays out this conceptual framework of components and roles. Later articles in this series will address the specific questions of software and information architectural design.Data Interoperability in Relation to Semantics
Semantic technologies give us the basis for understanding differences in meaning across sources, specifically geared to address differences in real world usage and context. These semantic tools are essential for providing common bases for relating structured data across various sources and contexts. These same semantic tools are also the basis by which we can determine what unstructured content “means”, thus providing the structured data tags that also enable us to relate documents to conventional data sources (from databases, spreadsheets, tables and the like). These semantic technologies are thus the key enablers for making information — unstructured, semi-structured and structured — understandable to both humans and machines across sources. Such understandings are then a key basis for powering the artificial intelligence applications that are now emerging to make our lives more productive and less routine.
For nearly a decade I have used an initial schema by Pluempitiwiriyawej and Hammer to elucidate the sources of possible semantic differences between content. Over the years I have added language and encoding differences to this schema. Most recently, I have updated this schema to specifically call out semantic heterogeneities due to either conceptual differences between sources (largely arising from schema differences) and value and attribute differences amongst actual data. I have further added examples for what each of these categories of semantic heterogenities means .
This table of more than 40 sources of semantic heterogeneities clearly shows the possible impediments to get data to interoperate across sources:Class Category Subcategory Examples Type   LANGUAGE Encoding Ingest Encoding Mismatch For example, ANSI v UTF-8  Concept Ingest Encoding Lacking Mis-recognition of tokens because not being parsed with the proper encoding  Concept Query Encoding Mismatch For example, ANSI v UTF-8 in search  Concept Query Encoding Lacking Mis-recognition of search tokens because not being parsed with the proper encoding  Concept Languages Script Mismatch Variations in how parsers handle, say, stemming, white spaces or hyphens Concept Parsing / Morphological Analysis Errors (many) Arabic languages (right-to-left) v Romance languages (left-to-right) Concept Syntactical Errors (many) Ambiguous sentence references, such as I’m glad I’m a man, and so is Lola (Lola by Ray Davies and the Kinks) Concept Semantics Errors (many) River bank v money bank v billiards bank shot Concept CONCEPTUAL Naming Case Sensitivity Uppercase v lower case v Camel case Concept Synonyms United States v USA v America v Uncle Sam v Great Satan Concept Acronyms United States v USA v US Concept Homonyms Such as when the same name refers to more than one concept, such as Name referring to a person v Name referring to a book Concept Misspellings As stated Concept Generalization / Specialization When single items in one schema are related to multiple items in another schema, or vice versa. For example, one schema may refer to “phone” but the other schema has multiple elements such as “home phone,” “work phone” and “cell phone” Concept Aggregation Intra-aggregation When the same population is divided differently (such as, Census v Federal regions for states, England v Great Britain v United Kingdom, or full person names v first-middle-last) Concept Inter-aggregation May occur when sums or counts are included as set members Concept Internal Path Discrepancy Can arise from different source-target retrieval paths in two different schemas (for example, hierarchical structures where the elements are different levels of remove) Concept Missing Item Content Discrepancy Differences in set enumerations or including items or not (say, US territories) in a listing of US states Concept Missing Content Differences in scope coverage between two or more datasets for the same concept Concept Attribute List Discrepancy Differences in attribute completeness between two or more datasets Attribute Missing Attribute Differences in scope coverage between two or more datasets for the same attribute Attribute Item Equivalence When two types (classes or sets) are asserted as being the same when the scope and reference are not (for example, Berlin the city v Berlin the official city-state) Concept When two individuals are asserted as being the same when they are actually distinct (for example, John Kennedy the president v John Kennedy the aircraft carrier) Attribute Type Mismatch When the same item is characterized by different types, such as a person being typed as an animal v human being v person Attribute Constraint Mismatch When attributes referring to the same thing have different cardinalities or disjointedness assertions Attribute DOMAIN Schematic Discrepancy Element-value to Element-label Mapping One of four errors that may occur when attribute names (say, Hair v Fur) may refer to the same attribute, or when same attribute names (say, Hair v Hair) may refer to different attribute scopes (say, Hair v Fur) or where values for these attributes may be the same but refer to different actual attributes or where values may differ but be for the same attribute and putative value.Many of the other semantic heterogeneities herein also contribute to schema discrepancies Attribute Attribute-value to Element-label Mapping Attribute Element-value to Attribute-label Mapping Attribute Attribute-value to Attribute-label Mapping Attribute Scale or Units Measurement Type Differences, say, in the metric v English measurement systems, or currencies Attribute Units Differences, say, in meters v centimeters v millimeters Attribute Precision For example, a value of 4.1 inches in one dataset v 4.106 in another dataset Attribute Data Representation Primitive Data Type Confusion often arises in the use of literals v URIs v object types Attribute Data Format Delimiting decimals by period v commas; various date formats; using exponents or aggregate units (such as thousands or millions) Attribute DATA Naming Case Sensitivity Uppercase v lower case v Camel case Attribute Synonyms For example, centimeters v cm Attribute Acronyms For example, currency symbols v currency names Attribute Homonyms Such as when the same name refers to more than one attribute, such as Name referring to a person v Name referring to a book Attribute Misspellings As stated Attribute ID Mismatch or Missing ID URIs can be a particular problem here, due to actual mismatches but also use of name spaces or not and truncated URIs Attribute Missing Data A common problem, more acute with closed world approaches than with open world ones Attribute Element Ordering Set members can be ordered or unordered, and if ordered, the sequences of individual members or values can differ Attribute Sources of Semantic Heterogeneities
Ultimately, since we express all of our content and information with human language, we need to start there to understand the first sources in semantic differences. Like the differences in human language, we also have differences in world views and experience. These differences are often conceptual in nature and get at what we might call differences in real world perspectives and experiences. From there, we encounter differences in our specific realms of expertise or concern, or the applicable domain(s) for our information and knowledge. Then, lastly, we give our observations and characterizations data and values in order to specify and quantify our observations. But the attributes of data are subject to the same semantic vagaries as concepts, in addition to their own specific challenges in units and measures and how they are expressed.
From the conceptual to actual data, then, we see differences in perspective, vocabularies, measures and conventions. Only by systematically understanding these sources of heterogeneity — and then explicitly addressing them — can we begin to try to put disparate information on a common footing. Only by reconciling these differences can we begin to get data to interoperate.
Some of these differences and heterogeneities are intrinsic to the nature of the data at hand. Even for the same putative topics, data from French researchers will be expressed in a different language and with different measurements (metric) than will data from English researchers. Some of these heterogeneities also arise from the basis and connections asserted between datasets, as misuse of the sameAs predicate shows in many linked data applications .
Fortunately, in many areas we are transitioning by social convention to overcome many of these sources of semantic heterogeneity. A mere twenty years ago, our information technology systems expressed and stored data in a multitude of formats and systems. The Internet and Web protocols have done much to overcome these sources of differences, what I’ve termed elsewhere as climbing the data federation pyramid . Semantic Web approaches where data items are assigned unique URIs are another source of making integration easier. And, whether all agree from a cultural aspect if it is good, we are also seeing English become the lingua franca of research and data.
The point of the table above is not to throw up our hands and say there is just too much complexity in data integration. Rather, by systematically decomposing the sources of semantic heterogeneity, we can anticipate and accommodate those sources not yet being addressed by cultural or technological conventions. While there is a large number of categories of semantic heterogeneity, these categories are also patterned and can be anticipated and corrected. These patterned sources inform us about what kind of work must be done to overcome semantic differences where they still reside.Work Components in Data Interoperability
The description logics that underly the semantic Web already do a fair job of architecting this concept-attribute split in semantics. The concept split is known as the TBox (for terminological knowledge, the basis for T in TBox) and represents the schema or taxonomy of the domain at hand. The TBox is the structural and intensional component of conceptual relationships. The second split of instances is known as the ABox (for assertions, the basis for A in ABox) and describes the attributes of instances (individuals), the roles between instances, and other assertions about instances regarding their class membership with the TBox concepts .
The semantic Web is a standards-based effort by the W3C (World Wide Web Consortium); many of its accomplishments have arisen around ontology and TBox-related efforts. Data integration has putatively been tackled from the perspective of linked data, but that methodology so far is short on attributes and property-mapping linkages between datasets and schema. There are as yet no reference vocabularies or schema for attributes . Many of the existing linked data linkages are based on erroneous owl:sameAs assertions. It is fair to say that attribute and ABox-level semantics and interoperability have received scarce attention, even though the logic underpinnings exist for progress to be made.
This lack on the attributes or ABox-side of things is a major gap in the work requirements for data interoperability, as we see from the table below. The TBox development and understanding is quite good; and, a number of reference ontologies are available upon which to ground conceptual mappings . But the ABox third is largely missing grounding references. And, the specialty work tasks, representing about the last third, are needful of better effectiveness and tooling.
For both the TBox and the ABox we are able to describe and model concepts (classes), instances (individuals), and are pretty good at being able to model relationships (predicates) between concepts and individuals. We also are able to ground concepts and their relationships through a number of reference concept ontologies . But our understanding of attributes (the descriptive properties of instances) remains poor and ungrounded. Best practices — let alone general practices — still remain to be discovered.TBox (concepts) Specialty Work Tasks ABox (data)
Across the knowledge base (that is, the combination of the TBox and the ABox), the semantic Web has improved its search capabilities by formally integrating with conventional text search engines, such as Solr. Instance and consistency checking are pretty straightforward to do, but are often neglected steps in most non-commercial semantic installations. Critical areas such as mappings, transformations and identity evaluation remain weak work areas. This figure helps show these major areas and their work splits:
As we discussed earlier on the recent and rapid advances of artificial intelligence , the combination of knowledge bases and the semantic Web with AI machine learning (ML) and NLP techniques will show rapid improvements in data interoperability. The two stumbling blocks of not having a framework and architecture for interoperability, plus the lack of attributes groundings, have been controlling. Now that these factors are known and they are being purposefully addressed, we should see rapid improvements, similar to other areas in AI.
This re-embedding of the semantic Web in artificial intelligence, coupled with the conscious attention to provide reference groundings for data interoperability, should do much to address what are current, labor-intensive stumbling blocks in the knowledge management workflow.Putting Some Grown-up Pants on the Semantic Web
The semantic Web clearly needs to play a central role in data integration and interoperability. Fortunately, like we have seen in other areas , semantic technologies lend themselves to generic functional software that can be designed for re-use in most any knowledge domain, chiefly by changing the data and ontologies guiding them. This means that reference libraries of groundings, mappings and transformations can be built over time and reused across enterprises and projects. Use of functional programming languages will also align well with the data and schema in knowledge management functions and ontologies and DSLs. These prospects parallel the emergence of knowledge-based AI (KBAI), which marries electronic Web knowledge bases with improvements in machine-learning algorithms.
The time for these initiatives is now. The complete lack of distributed data interoperability is no longer tolerable. High costs due to unacceptable manual efforts and too many failed projects plague the data interoperability efforts of the past. Data interoperability is no longer a luxury, but a necessity for enterprises needing to compete in a data-intensive environment. At scale, point-to-point integration efforts become ineffective; a form of reusable and transferable master data management (MDM) needs to emerge for the realiites of Big Data, and one that is based on the open and standard protocols of the Web.
Much tooling and better workflows and user interfaces will need to emerge. But the critical aspects are the ones we are addressing now: information and software architectures; reference groundings and attributes; and education about these very real prospects near at hand. The challenge of data interoperability in cooperation with its artificial intelligence cousin is where the semantic Web will finally put on its Big Boy pants. See Charnyote Pluempitiwiriyawej and Joachim Hammer, 2000. A Classification Scheme for Semantic and Schematic Heterogeneities in XML Data Sources, Technical Report TR00-004, University of Florida, Gainesville, FL, 36 pp., September 2000. See https://cise.ufl.edu/tr/DOC/REP-2000-396.pdf. I first cited this report and extended it to cover languages (see ) in M.K. Bergman 2006. Sources and Classification of Semantic Heterogeneities, AI3:::Adaptive Information blog, June 6, 2006. See http://www.mkbergman.com/232/sources-and-classification-of-semantic-heterogeneities/). This most recent version added the examples and expanding the listing a bit further, to where it is no longer faithful to the original 2000 paper.  Concept is the shorthand used for the schema or classes or TBox. Attribute is the shorthand used for instance data or entities and their ABox. I segregate class-relation properties (predicates) from instance-describing properties (attributes). This distinction is not use in standard TBox-ABox splits; its rationale will be described in a further article.  See M.K. Bergman, 2006. Tutorial: Internet Languages, Character Sets and Encodings, BrightPlanet Corporation Technical Documentation, March 2006, 13 pp. See http://www.mkbergman.com/wp-content/themes/ai3v2/files/2006Posts/InternationalizationTutorial060323.pdf.  See . Also the TBox portion, or classes (concepts), is the basis of the ontologies. The ontologies establish the structure used for governing the conceptual relationships for that domain and in reference to external (Web) ontologies. The ABox portion, or instances (named entities), represents the specific, individual things that are the members of those classes. Named entities are the notable objects, persons, places, events, organizations and things of the world. Each named entity is related to one or more classes (concepts) to which it is a member. Named entities do not set the structure of the domain, but populate that structure. The ABox and TBox play different roles in the use and organization of the information and structure.  M.K. Bergman 2009. When Linked Data Rules Fail, AI3:::Adaptive Information blog, November 16, 2009. See http://www.mkbergman.com/846/when-linked-data-rules-fail/.  M.K. Bergman 2006. Climbing the Data Federation Pyramid, AI3:::Adaptive Information blog, May 25, 2006. See http://www.mkbergman.com/229/climbing-the-data-federation-pyramid/.  M.K. Bergman 2008. Thinking ‘Inside the Box’ with Description Logics, AI3:::Adaptive Information blog, November 10, 2008. See http://www.mkbergman.com/466/thinking-inside-the-box-with-description-logics/.  See the thread on the W3C semantic web mailing list beginning at http://lists.w3.org/Archives/Public/semantic-web/2014Jul/0129.html.  Examples of upper-level ontologies include UMBEL, the Suggested Upper Merged Ontology (SUMO), the Descriptive Ontology for Linguistic and Cognitive Engineering (DOLCE), PROTON, Cyc and BFO (Basic Formal Ontology). Most of the content in their upper-levels is akin to broad, abstract relations or concepts (similar to the primary classes, for example, in a Roget’s Thesaurus) than to “generic common knowledge.” Most all of them have both a hierarchical and networked structure, though their actual subject structure relating to concrete things is generally pretty weak. See further the Wikipedia entry on upper ontologies.  M.K. Bergman 2014. Spring Dawns on Artificial Intelligence, AI3:::Adaptive Information blog, June 2, 2014. See http://www.mkbergman.com/1731/spring-dawns-on-artificial-intelligence/.  M.K. Bergman 2011. Ontology-driven Apps Using Generic Applications, AI3:::Adaptive Information blog, March 7, 2011. See http://www.mkbergman.com/948/ontology-driven-apps-using-generic-applications/.
In our recent two-part series we described a decade of experience working in the semantic Web (Part I) and our view that Big Structure, which resides at the nexus of the semantic Web, knowledge bases and artificial intelligence, was a key component of making sense of Big Data going forward (Part II). We are at a time when multiple advances are conjoining to create new opportunities and excitement.
Data without context and relationships is meaningless. The idea of Big Data is powerful, but it is often presented as either a “good thing” in and of itself, or a mantra for something that is rather undefined. There is no doubt that with the Internet and the Web we are now able to generate and access data at unprecedented scale. There is also no question that tracking mechanisms and cheap storage — and simpler, large-scale databases and Web services — mean that we can also capture data and structure of natures previously unseen. Everyone knows the remarkable growth in exabytes and more.
The prospect of data everywhere — some useful with important context and some not — has clearly captured the current discussion. Heck, if we claim Big Data, we even make more in wage or consulting charge-out fees. Who can argue with that?
Well, actually, anyone interested in meaningful data or cross-dataset interoperability can argue with that. Big Data is great, except it means little if we can not combine that data across multiple sources for potentially multiple purposes. (Remember, one of the “V’s” of Big Data is variability.) Once the question of what data means gets brought to the fore, it is now time for context and relationships. Structure in an information context means that which situates or describes data in an interpretable way. Big Data needs a Big Structure complement to make sense of it all.What is a Big Structure?
Big Structure is data relationships and context that can be combined into a coherent framework to enable dataset interoperability and understanding. By necessity, Big Structure implies that the meaning of data can be understood and its values can be brought to common bases such that analysis, testing and validation can be applied across values. Big Structure is not a monolithic thing, but the combination of multiple things that give data meaning and context. As such, Big Structure is often a re-purposing of existing structural assets, plus other special sauce, organized for the aim of data interoperability.Big Structure is data relationships and context that can be combined into a coherent framework to enable dataset interoperability and understanding.
The components of Big Structure can be identified and characterized. These components can be assessed for usefulness and authoritativeness, and then incorporated into broader structures that ultimately bring the topics of what the data is about and the values of that data into alignment. Thus, Big Structure is also a mindset and approach to selecting and combining structures such that broad dataset interoperability can be achieved.
Big Structure is actually a continuum or family of concept and data relationships, any one of which is also a contributor to helping to map and interoperate data. Ultimately, the components of Big Structure get combined into reference graph structures that place the concepts and actual data values of the Big Data into context. There are certain ways to use and organize existing structures to achieve these Big Structure objectives; some of these ways are described in this article.
Once the components of Big Structure are combined into these reference graphs we then can also use network or graph analysis to understand the relationships amongst the constituent data items. This recursive nature of graph reference structures to organize the constituent data and then to use those graphs to analyze the data is one of the hallmark characteristics of Big Structure.
Big Structure thus involves the need to identify and then organize constituent forms of structure into coherent reference frameworks. Concepts in contributing datasets are then mapped to these structures, and the attributes and values of the underlying data are also transformed into canonical representations. It is these mappings and transformations that provide the interoperability of Big Structure. Big Structure therefore continues to evolve by adding more and more reference structures, all coherently organized.Contributors to Big Structure
Big Structure is a family of canonical reference structures that help guide mapping and interoperability. The table below lists some of the possible contributors to Big Structure , roughly in descending order as to the degree of structure and its contribution to interoperability. The table provides both definitions and use descriptions for each component, plus optionally some notes regarding coverage and use:Structure Type Definition Use Note Reference ontologies Major grounding structures for orienting and interoperating concepts or data The reference concepts for orienting all data and domain information  Reference attributes Major grounding structures for interoperating data and data characterizations The reference relationships amongst data descriptions and characteristics, which also provides the means for transformations between heterogeneous representations  Data model (RDF) A self-consistent means for describing the structure of data and their relationships The “canonical” data model at the heart of the system; provides a single interoperability point; RDF is the canonical model used by Structured Dynamics for its Big Structures  Domain attributes The data descriptions and characteristics for the constituent datasets in the applicable domain(s) The reference attributes specific to the domain(s) at hand (which are generally more specific than general reference attributes) Domain ontologies The formal conceptualization of a domain, using a shared vocabulary to denote the types, properties and interrelationships of those concepts The reference concepts and their relationships specific to the domain(s) at hand; generally are mapped to the reference ontologies  Concept maps A diagram that depicts suggested relationships between concepts Structurally similar to a domain ontology; a few related terms shown in Note  Schema The structure of a database that defines the objects and relationships in that database Organizing framework for relational databases (and their tables)  Mappings The process of creating data element correspondences between two distinct data models or schema Mapping predicates are used to relate concepts or attributes from two different datasets or knowledge bases to one another. Mappings are often a precursor to various transformations to bring data into a common representation  Taxonomies A particular classification of related concepts, often of a hierarchical nature Hierarchical relationships are expressed in narrower or broader terms (or subClassOf); may also be see also relationships  Facets Clearly defined, mutually exclusive, and collectively exhaustive aspects, properties or characteristics of a class or specific subject Facets can provide alternative ways for classifying objects beyond a single taxonomy Categories Grouping objects based on similar properties A category may be viewed as equivalent to a concept  Tables A collection of related data held in a structured format, generally a two-dimensional layout of rows (records) and columns (fields) Simplest and most common data presentation format Synsets A group of data elements or terms that are considered semantically equivalent for the purposes of information retrieval Also known as a “semset” in the parlance of UMBEL Metadata Data providing information about one or more aspects of the source data, thus “data about data” It is the description of what data is about rather than the values and attributes of the actual data Thesauri A form of controlled vocabulary that seeks to dictate semantic manifestations of metadata in the indexing of content objects A thesaurus is composed a list of words (or terms), a vocabulary for relating these words (or terms) to one another, often hierarchically, and a set of rules on how to use these aspects Gazetteers A listing of similar entity types with associated structural data (such as countries and population or standard codes) Often used in relation to people or place entity types, though any class of entities may have a gazetteer Controlled vocabularies The use of predefined, authorized terms as preselected by the sponsor to enforce consistency in terminology Applied to specific domains or sub-domains, with single controlled vocabularies per official language used Reference lists Authoritative listings of similar objects, each uniquely identified by name or code May be as simple as a comprehensive list of countries with associated ISO codes  Dictionaries A repository of information about data such as meaning, relationships to other data, origin, usage, or format In our context, can range from the meaning associated with standard word dictionaries to the more formal data dictionary Glossaries An alphabetical list of terms in a particular domain with the definitions for those terms Definition is the only structured information provided Nested lists Related concepts or entities organized by some form of hierarchical relationship (narrower, broader, subClassOf, etc.) Akin to a simple taxonomy Ordered lists A finite, ordered collection of values for a given type May also be additional information linked to the listing Clusters A set of objects grouped according to some basis of similarity (type, attributes, or characteristics) Basis for how the objects got clustered is not always obvious Unordered lists A container of similar items or entities, with no implied order or sequence Also known as a “bag” or “collection”  Values The actual data; a normal form or a type member Basic QUDT ontologies could contribute here
An alternate way to look at these contributor structures is to characterize them with respect to degree of structure and degree of contributing to interoperability:
In general, as might be expected, the greater the degree of structure, the greater its potential contribution to interoperability. The components in the upper right quadrant represent the most structured and interoperable ones. These also conform most to the use of W3C standards for the RDF data model and the OWL ontology languages. Expressions of structure are codified and standardized. Use of best practices also ensures completeness and suitability as reference groundings for interoperability.
The lower left portions of the quadrant represent the least structure and interoperability. However, as standard reference means for characterizing and describing data, even structures in this quadrant can contribute to meeting Big Structure requirements. Tagging of documents (unstructured data) occurs in this less-sophisticated lower left quadrant, but it gives equal footing to 80% of the content that generally resides in text form. (The interoperability system is further enhanced when the basis of the tags is derived from the “semsets” of the reference and domain ontologies, another example of a best practice.)
All of the listed components can thus contribute to Big Structure. However, the completeness of that structure and its usefulness for interoperability increases as one progresses along the blue arrow of the Big Structure continuum. Data interoperability arises from the continued efforts to drive Big Structure to the upper right of this quadrant. As noted, Big Structure is a mindset and process rather than some finite state. As more concepts and attributes get grounded in standard references, the degree of Big Structure (and, thus, data interoperability) continues to increase.The Foundation of Reference Groundings
In both semantics and artificial intelligence — and certainly in the realm of data interoperability — there is always the problem of symbol grounding. In the conceptual realm, symbol grounding means that when we use a term or phrase we are referring to the same thing; that is, the referent is the same. In the data value realm, symbol grounding means that when we refer to an object or a number — say, the number 4.1 — we are also referring to the same metric. 4.1 inches is not the same as 4.1 centimeters or 4.1 on the Richter scale, and object names for set member types also have the same challenges of ambiguous semantics as do all other things referred to by language.
The variability V in Big Data or the 40-some dimensions of potential semantic heterogeneity  are explicit recognitions of the symbol grounding challenge. Assuming we can determine context (itself an important consideration not further discussed here), fixity of reference is essential to these groundings. Context and groundings are the ways by which we remove ambiguity in what we measure and record.
Like dictionaries for human languages, or stars and constellations for navigators, or agreed standards in measurement, or the Greenwich meridian for timekeepers, fixed references are needed to orient and “ground” each new dataset over which we attempt to integrate. Without such fixities of reference, everything floats in reference to other things, the cursed “rubber ruler” phenomenon.
Thus, we can express our Big Structure components from a foundational perspective as well. In Structured Dynamics‘ view of the world, the foundation for data interoperability is grounded in reference structures or ontologies that provide the fixity of reference for concepts and data and their attributes. Upon these foundations are then constructed the domain views of concepts and attributes, which become the target for mapping other references and Big Structures:
The mappings, transformations and domain and reference ontologies are themselves written in the OWL languages of the W3C and the standards of the RDF data model. At this most expressive end of Big Structure, the representations are in the form of graphs. Network and graph analytics will expand still further business intelligence prospects. The use of these standards with common and testable logic is another means to ensure coherency and interoperability of the Big Structure that results.
Note a key aspect of the grounding foundation is missing: one or more reference ontologies for attributes. Though many examples exist on the concept side, little has been done to explicitly address the questions of data value interoperability. This major gap is a current emphasis of Structured Dynamics, with much that will be said over the coming weeks. Also expect an open source reference ontology for attributes in the near future.
The thing is that we are learning how to make the various parts of this interoperability stack work. We are leveraging existing structural assets of all kinds to establish the semantics and infrastructure for domain interoperability. We know how to match and map these existing structural assets to the reference frameworks that are the foundation to interoperability.A Vision of Interoperability
The real world is one of heterogeneous datasets, multiple schema and differing viewpoints. Even within single enterprises — and those which formerly expressed little need or interest to interoperate with the broader world — data integration and interoperability has been a real challenge. Big Data itself is not solving these problems. Quite the opposite. Big Data trends are turning data interoperability molehills into mountain-high competitive threats.
Like any well-built structure, data interoperability requires a solid foundation. That foundation must reside in exemplar reference ontologies upon which to ground the semantics and exchange standards for data. Using the canonical RDF data model makes this task practical. Existing information structures of various types across the enterprise and the Web all can and should play a role in establishing reference structures. The accretion of reference structures will lead to still further interoperability and the ability to incorporate more datasets. Currently expensive practices in, say, master data management (MDM) can begin to transition to a new paradigm. It is easy to envision working from a library of existing reference standards for use across enterprises. This kind of incremental expansion of interoperability leads to still more interoperable data in a virtuous cycle of innovation and lower budgets.
As our computing continues to get more virtual and cloud-like, physical and hardware and software architectures must give way to information architectures (in the true sense of interoperability). We have no choice but to treat the architecting of information as a first-order challenge. The totally cool thing about the data integration challenge is that the architecture can be readily varied and tested to achieve a working foundation. Much empirical information exists about how to do it and what to do next. The chief challenge has been to recognize that data interoperability — and its dependence on Big Structure — is a first-order concern (and opportunity). The intersection of Big Structure with Big Data, and with graph and AI algorithms, should create new approaches to chew across the data integration environment. I expect progress to be rapid. There are at least 40 terms or concepts across these various disciplines, most related to Web and general knowledge content, that have organizational or classificatory aspects that — loosely defined — could be called an “ontology” framework or approach. See M.K. Bergman, 2007. An Intrepid Guide to Ontologies, AI3:::Adaptive Information blog, May 16, 2007.  UMBEL and other upper level ontologies are examples here. In the case of UMBEL, that Big Structure is used as a scaffolding of reference concepts used to link external (unrelated) structures to help inter-operating data between two unrelated systems. Such a Big Structure can also be used for other tasks such as helping machine learning techniques to categorize and disambiguate pieces of data by leveraring such a structure of types.  Unfortunately, no reference structures for attributes yet exist. For a discussion of this status, see the thread on the W3C semantic web mailing list beginning at http://lists.w3.org/Archives/Public/semantic-web/2014Jul/0129.html.  Data models encompass a rather broad span. The RDF discussion represents a more formal end of the data model spectrum, wherein there is complete logic, syntax and serialization discussions, more involved than most data models.  Domain ontologies represent the most closely-aligned view of the domain and its relationships of all of the component structures listed.  Concept maps are very closely related to ontologies, and may include topic maps, mind maps and other graph-like structures of concepts.  Schema may apply to many realms, but in the IT and software context schema mostly refers to database schema related to relational databases. These are often expresssed in UML diagrams or XML schema.  Mappings and transformatons are a huge area of diverse structure and different serializations and specifications. Fortunately, the task of mapping external structure to RDF removes the many-to-many issues with most transformation approaches.  Taxonomies mask an entire sub-categories of directories, folksonomies, subject trees, and more. The key aspect is that relevant concepts are expressed in a graph relationship manner to other concepts, often in a hierarchical fashion.  Categories also includes the general classification process.  I would consider a canonical references listing of country names and codes to be a part of Big Structure, since they act as a controlled vocabulary.  This is a key area for including unstructured documents, since tags are a primary means of adding metadata to a document. When the pool of tags is based on the governing reference and domain ontologies, then interoperability is further promoted.  M.K. Bergman, 2006. Sources and Classification of Semantic Heterogeneities, AI3:::Adaptive Information blog, June 6, 2006.
In Part I of this two-part series, Fred Giasson and I looked back over a decade of working within the semantic Web and found it partially successful but really the wrong question moving forward. The inadequacies of the semantic Web to date reside in its lack of attention to practical data interoperability across organizational or community boundaries. An emphasis on linked data has created an illusion that questions of data integration are being effectively addressed. They are not.
Linked data is hard to publish and not the only useful form for consuming data; linked data quality is often unreliable; the linking predicates for relating disparate data sources to one another may be inadequate or wrong; and, there are no reference groundings for relating data values across datasets. Neither the semantic Web nor linked data has developed the practices, tooling or experience to actually interoperate data across the Web. These criticisms are not meant to condemn linked data — it is, after all, the early years. Where it is compliant and from authoritative information sources, linked data can be a gold standard in data publishing. But, linked data is neither necessary nor essential, and may even be a diversion if it sucks the air from the room for what is more broadly useful.
This table summarizes the state-of-art in the semantic Web for frameworks and guidance in how to interoperate data:Category Related Terms Status in the Semantic Web Notes Classes sets, concepts, topics, types, kinds Mature, but broader scope coverage desirable; equivalent linkages between datasets often mis-applied; more realistic proximate linkages in flux, with no bases to reason over them  Instances individuals, entities, members, records, things Current basis for linked data; many linkage properties mis-applied  Relation Properties relations, predicates Equivalent linkages between datasets often mis-applied; more realistic proximate linkages in flux, with no bases to reason over them.  Descriptive Properties attributes, descriptors Save for a couple of minor exceptions, no basis for mapping attributes across datasets  Values data Basic QUDT ontologies could contribute here 
We can relate the standard subject – predicate – object triple statement in RDF to this table, using the Category column. Classes and Instances relate to the subjects, Relation and Descriptive Properties relate to the predicate, and Values relate to the object  in an RDF triple. The concepts and class schema of different information sources (their “aboutness”) can reasonably be made to interoperate. In terms of the description logics that underly the logic bases of W3C ontologies, the focus and early accomplishments of the semantic Web have been on this “terminological box” or T-Box . Tooling to make the mappings more productive and means to test the coherence and completeness of the results still remain as priority efforts, but the conceptual basis and best practices have progressed pretty well.
In contrast, nearly lacking in focus and tooling has been the flip side of that description logics coin: the A-Box , or assertional and instance (data) level of the equation. Both the T-Box and A-Box are necessary to provide a knowledge base. Today, there are virtually no vocabularies, no tooling, no history, no best practices and no “grounding” for actual A-Box data integration within the semantic Web. Without such guidance, the semantic Web is silent on the questions of data interoperability. As David Karger explained in his keynote address at ISWC in 2013 , “we’ve got our heads in the clouds while people are stuck in the dirt.”
Yet these are not fatal flaws of the semantic Web, nor are they permanent. Careful inspection of current circumstances, combined with purposeful action, suggests:
Why do we keep pointing to the question of data interoperability? Consider these facts:
The abiding, costly, frustrating and energy-sucking demands of data integration have been a constant within enterprises for more than three decades. The same challenges reside for the Web. The Internet of Things will further demand better interoperability frameworks and guidelines. Current data integration tooling relies little upon semantics and no leading alternative is based principally around semantic approaches .
The data integration market is considered to include enterprise data integration and extract, transform and load (ETL) vendors. Gartner estimates tool sales for this market to be about $2 billion annually, with a growth rate faster than most IT areas . But data integration also touches upon broader areas such as enterprise application integration (EAI), federated search and query, and master data management (MDM), among others. Given that data integration is also 40% of standard IT project costs, new approaches are needed to finally unblock the costly logjam of enterprise information integration. Most analysts see firms that are actively pursuing data integration innovations as forward-thinking and more competitive.
Data integration is combining information from multiple sources and providing users a uniform view of it. Data interoperability is being able to exchange and work upon (inter-operate) information across system and organizational boundaries. The ability to integrate data precedes the ability to interoperate it. For example, I may have three datasets of mammals that I want to consolidate and describe in similar terms with common units of measurement. That is an example of data integration. I may then want to relate this mammal knowledge base with a more general perspective of the animal kingdom. That is an example of data interoperability. Data integration usually occurs within a single organization or enterprise or institutional offering (as would be, say, Wikipedia). Data interoperability additionally needs to define meanings and communicate them in common ways across organizational, domain or community boundaries.
These are natural applications for the semantic Web. Why, then, has there not been more practical use of the semantic Web for these purposes?
That is an interesting question that we only partially addressed in Part I of this series. All aspects of data have semantics: what the data is about, what is its context, how it relates to other data, and what its values are and what they mean. The semantic Web is closely allied with natural language processing, an essential for bringing the 80% of unstructured data into the equation. Semantic Web ontologies are useful structures for how to relate real-world data into common, reference forms. The open world logic of the semantic Web is the right perspective for knowledge functions under the real-world conditions of constantly expanding information and understandings.
While these requirements suggest an integral role for the semantic Web, it is also clear that the semantic Web has not yet made these contributions. One explanation may be that semantic Web advocates, let alone the linked data tribe, have not seen data integration — as traditionally defined — as their central remit. Another possibility is that trying to solve data interoperability through the primary lens of the semantic Web is the wrong focus. In any case, meeting the challenge of data interoperability clearly requires a much broader context.Embedding Data Interoperability Into a Broader Context
The semantic Web, in our view, is properly understood as a sub-domain of artificial intelligence. Semantic technologies mesh smoothly with natural language tasks and objectives. But, as we noted in a recent review article, artificial intelligence is itself undergoing a renaissance . These advances are coming about because of the use of knowledge-based AI (KBAI), which combines knowledge bases with machine learning and other AI approaches. Natural language and spoken interfaces combined with background knowledge and a few machine-language utilities are what underlie Apple’s Siri, for example.
The realization that the semantic Web is useful but insufficient and that AI is benefitting from the leveraging of background knowledge and knowledge bases caused us to “decompose” the data-interoperability information space. Because artificial intelligence is a key player here, we also wanted to capture all of the main sub-domains of AI and their relationships to one another:Artificial Intelligence Domains
Two core observations emerge from standing back and looking at these questions. First, many of AI’s main sub-domains have a role to play with respect to data integration and interoperability:AI Domains Related to Data Interoperability
This places semantic Web technologies as a co-participant with natural language processing, knowledge mining, pattern recognizers, KR languages, reasoners, and machine learning as domains related to data interoperability.
And, second, generalizing the understanding of knowledge bases and other guiding structures in this space, such as ontologies, highlights the potential importance of Big Structure. Virtually every one of the domains displayed above would be aided by leveraging background knowledge.Grounding Data Interoperability in Big Structure
As our previous AI review showed , reference knowledge bases — Wikipedia in the forefront — have been a tremendous boon to moving forward on many AI challenges. Our own experience with UMBEL has also shown how reference ontologies can help align and provide common grounding for mapping different information domains into one another . Vetted, gold-standard reference structures provide a fixity of coherent touchpoints for orienting different concepts and domains (and, we believe, data) to one another.
In the data integration context, master data models (and management, or MDM) attempt to provide common reference terms and objects to aid the integration effort. Like other areas in conventional data integration, very few examples of MDM tools based on semantic technologies exist.
This use of reference structures and the importance of knowledge bases to help solve hard computational tasks suggests there may be a general principle at work. If ontologies can help orient domain concepts, why can’t they also be used to orient instance data and their attributes? In fact, must these structures always be ontologies? Are not other common reference structures such as taxonomies, vocabularies, reference entity sets, or other typologies potentially useful to data integration?
By standing back in this manner and asking these broader questions we can see a host of structures like reference concepts, reference attributes, reference places, reference identifiers, and the like, playing the roles of providing common groundings for integration and interoperation. Through the AI experience, we can also see that subsequent use of these reference structures — be they full knowledge bases or more limited structures like taxonomies or typologies — can further improve information extraction and organization. The virtuous circle of knowledge structures improving AI algorithms, which can then further improve the knowledge structures, has been a real Aha! moment for the artificial intelligence community. We should see rapid iterations of this virtuous circle in the months to come.
These perspectives can help lead to purposeful designs and approaches for attacking such next-generation problems as data interoperability. The semantic Web can not solve this alone because additional AI capabilities need to be brought to bear. Conventional data integration approaches that lack semantic Big Structure groundings — let alone the use of AI techniques — have years of history of high cost and disappointing results. No conventional enterprise knowledge management problem appears sheltered from this whirlwind of knowledge-backed AI.
At Structured Dynamics, Fred Giasson and I have been discussing “Big Structure” for some time. However, it was only in researching this article that I came across the first public use of this phrase in the context of AI and big data. In May, Dr. Jiawei Han, a leading researcher in data mining, gave a lecture at Yahoo! Labs entitled, Big Data Needs Big Structure. In it, he defines “Big Structure as a type information network.” The correlation with ontologies and knowledge structures is obvious.An Emerging Development Agenda
The intellectual foundations already exist to move aggressively on a focused development agenda to improve the infrastructure of data interoperability. This emerging agenda needs to look to new refererence structures, better tooling, the use of functional languages and practices, and user interfaces and workflows that improve the mappings that are the heart of interoperability.
Big Structure, such as UMBEL for referencing what data is about, is the present exemplar for going forward. Excellent reference and domain ontologies for common domains already exist. Mapping predicates have been developed for these purposes. Though creation of the maps is still laborious, tooling improvements (see below) should speed up that process as well.
What is next needed are reference structures to help guide attributes mappings, data value mappings, and transformations into usable common attribute quantities and types. I will discuss in a later post our more detailed thoughts of what a reference gold-standard attribute ontology should look like. This new Big Structure should also be helpful in guiding conversion, transformation and “lifting” utilities that may be used to bring attribute values from heterogeneous sources into a common basis. As mappings are completed, these too can become standard references as the bootstrapping continues.
Mappings for data integration across the scales, scope and growth of data volumes on the Web and within enterprises can no longer be handled manually. Semi-automated tooling must be developed and refined that operates over large volumes with acceptable performance. Constant efforts to reduce the data volumes requiring manual curation are essential; AI approaches should be incorporated into the virtuous iterations to reduce these efforts. Meanwhile, attentiveness to productive user interfaces and efficient workflows are also essential to improve throughput.
Further, by working off of standards-based Big Structures, this tooling can be made more-or-less generic, with ready application to different domains and different data. Because this tooling will often work in enterprises behind firewalls, standard enterprise capabilities (security, access, preservation, availability) should also be added to this infrastructure.
These Big Structures and tools should themselves be created and maintained via functional programming languages and DSLs specifically geared to the circumstances at hand. We want languages suited to RDF and AI purposes with superior performance across large mapped datasets and unstructured text. But we also want languages that are easier to use and maintain by knowledge workers themselves. Partitioning strategies may also need to be employed to ensure acceptable real-time feedback to users responsible for data integration mappings.A New Adaptive Infrastructure for Data Interoperability
Structured Dynamics’ review exercise, now documented in this two-part series, affirms the semantic Web needs to become re-embedded in artificial intelligence, backed by knowledge bases, which are themselves creatures of the semantic Web. Coupling artificial intelligence with knowledge bases will do much to improve the most labor-intensive stumbling blocks in the data integration workflow: mappings and transformations. Through a purposeful approach of developing reference structures for attributes and data values, we will begin to see marked improvements in the efficiency and lower costs of data integration. In turn, what is learned by using these approaches for mastering MDM will teach the semantic Web much.
An approach using semantic technologies and artificial intelligence tools will begin to solve the data integration puzzle. By leveraging background knowledge, we will begin to extend into data interoperability. Purposeful attention to tooling and workflows geared to improve the mapping speed and efficiency by users will enable us to increase the stable of reference structures — that is, Big Structure — available for the next integration challenges. As this roster of Big Structures increases, they can be shared, allowing more generic issues of data integration to be overcome, freeing domains and enterprises to target what is unique.
Achieving this vision will not occur overnight. But, based on a decade of semantic Web experience and the insights being gained from today’s knowledge-based AI advances, the way forward looks pretty clear. We are entering a fundamental new era of knowledge-based computation. We welcome challenging case examples that will help us move this vision forward.NOTE: This Part II concludes the series with Part I, A Decade in the Trenches of the Semantic Web  Using semantic ontologies can and has worked well for many domains and applications, such as the biomedical OBO ontologies, IBM’s Watson, Google’s Knowledge Graph, and hundreds in more specific domains. Combined with concept reference structures like UMBEL, both building blocks and exemplars exist for how to interoperate across what different domains are about.  For examples of issues, see M. K. Bergman, 2009. When Linked Data Rules Fail, AI3:::Adaptive Information blog, November 16, 2009.  Some of these options are overviewed by M. K. Bergman, 2010. The Nature of Connectedness on the Web, AI3:::Adaptive Information blog, November 22, 2010.  See the thread on the W3C semantic web mailing list beginning at http://lists.w3.org/Archives/Public/semantic-web/2014Jul/0129.html.  See QUDT – Quantities, Units, Dimensions and Data Types Ontologies, Retrieved July 22, 2014.  The object may also refer to another class or instance, in which case the relation property takes the form of an ObjectProperty and the “value” is the URI referring to that object.  See, for example, M. K. Bergman, 2009. Making Linked Data Reasonable Using Description Logics, Part 2, AI3:::Adaptive Information blog, February 15, 2009.  See David Karger, 2013. Keynote at the European Semantic Web Conference Part 1: The State of End User Information Management, June 5, 2013.  Info-Tech Research Group, 2011. Vendor Landscape Plus: Data Integration Tools, 72 pp.  Gartner estimates that the data integration tool market was slightly over $2 billion at the end of 2012, an increase of 7.4% from 2011. This market is seeing an above-average growth rate of the overall enterprise software market, as data integration continues to be considered a strategic priority by organizations. See Eric Thoo, Ted Friedman, Mark A. Beyer, 2013. Magic Quadrant for Data Integration Tools, research Report G00248961 from Gartner, Inc., 17 July 2013; see: http://www.gartner.com/technology/reprints.do?id=1-1HBEFSF&ct=130717&st=sb  See M. K. Bergman, 2014. Spring Dawns on Artificial Intelligence, AI3:::Adaptive Information blog, June 2, 2014.  See M. K. Bergman, 2011. In Search of ‘Gold Standards’ for the Semantic Web, AI3:::Adaptive Information blog, February 28, 2011.
Cinemaphiles will readily recognize Akira Kurosawa‘s Rashomon film of 1951. And, in the 1960s, one of the most popular book series was Lawrence Durrell‘s The Alexandria Quartet. Both, each in its own way, tried to get at the question of what is truth by telling the same story from the perspective of different protagonists. Whether you saw this movie or read these books you know the punchline: the truth was very different depending on the point of view and experience — including self-interest and delusion — of each protagonist. All of us recognize this phenomenon of the blind men’s view of the elephant.
I have been making my living and working full time on the semantic Web and semantic technologies now for a full decade. So has my partner at Structured Dynamics, Fred Giasson. Others have certainly worked longer in this field. The original semantic Web article appeared in Scientific American in 2000 , and the foundational Resource Description Framework data model dates from 1999. Fred and I have our own views of what has gone on in the trenches of the semantic Web over this period. We thought a decade was a good point to look back, share what we’ve experienced, and discover where to point our next offensive thrusts.What Has Gone Well?
The vision of the semantic Web in the Scientific American article painted a picture of globally interconnected data leveraged by agents or bots designed to make our lives easier and more automated. However, by the time that I got directly involved, nearly five years after standards first started to be published, Tim Berners-Lee and many leading proponents of RDF were beginning to shift focus to linked data. The agents, and automation, and ontologies of the initial vision were being downplayed in favor of effective means to publish and consume data based on RDF. In many ways, linked data resembled a re-branding.
This break had been coming for a while, memorably captured by a 2008 ISWC session led by Peter F. Patel-Schneider . This internal division of viewpoint likely caused effort to be split that would have been better spent in proselytizing and improving tools. It also diverted somewhat into internal squabbles. While many others have pointed to a tactical mistake of using an XML serialization for early versions of RDF as a key factor is slowing initial adoption, a factor I agree was at play, my own suspicion is that the philosophical split taking place in the community was the heavier burden.
Whatever the cause, many of the hopes of the heady days of the initial vision have not been obtained over the past fifteen years, though there have been notable successes.
The biomedical community has been the shining exemplar for data interoperability across an entire discipline, with earth sciences, ecology and other science-based domains also showing interoperability success . Families of ontologies accompanied by tooling and best practices have characterized many of these efforts. Sadly, though, most other domains have not followed suit, and commercial interoperability is nearly non-existent.
Most all of the remaining success has resided in single-institution data integration and knowledge representation initiatives. IBM’s Watson and Apple’s Siri are two amazing capabilities run and managed by single institutions, as is Google’s Knowledge Graph. Also, some individual commercial and government enterprises, willing to pay support to semantic technology experts, have shown success in data integration, using RDF, SKOS and OWL.
We have seen the close kinship between natural language, text, and Q & A with the semantic Web, also demonstrated by Siri and more recent offshoots. We have seen a trend toward pairing great-performing open source text engines, notably Solr, with RDF and triple stores. Recommendation systems have shown some success. Linked data publishing has also had some notable examples, including the first of the lot, DBpedia, with certain institutional publishers (such as the Library of Congress, Eurostat, The Getty, Europeana, OpenGLAM [galleries, archives, libraries, and museums]) showing leadership and the commitment of significant vocabularies to linked data form.
On the standards front, early experience led to new and better versions of the SPARQL query language (SPARQL 1.1 was greatly improved in the last decade and appears to be one capability that sells triple stores), RDF 1.1 and OWL 2. Certain open source tools have become prominent, including Protégé, Virtuoso (open source) and Jena (among unnamed others, of course). At least in the early part of this history, tool development was rapid and flourishing, though the innovation pace has dropped substantially according to my tracking database Sweet Tools.What Has Disappointed?
My biggest disappointments have been, first, the complete lack of distributed data interoperability, and, second, the lack or inability of commercial enterprises to embrace and adopt semantic technologies on their own. The near absence of discussion about instance records and their attributes helps frame the current maturity of the semantic Web. Namely, it has yet to crack the real nuts of data integration and interoperability across organizations. Again, with the exception of the biomedical community, neither in the linked data realm nor in the broader semantic Web, can we point to information based on semantic Web principles being widely shared between systems and organizations.
Some in the linked data community have explicitly acknowledged this. The abstract for the upcoming COLD 2014 workshop, for example, states :. . . applications that consume Linked Data are not yet widespread. Reasons may include a lack of suitable methods for a number of open problems, including the seamless integration of Linked Data from multiple sources, dynamic discovery of available data and data sources, provenance and information quality assessment, application development environments, and appropriate end user interfaces.
We have written about many issues with linked data, ranging from the use of improper mapping predicates; to the difficulty in publishing; and to dereferencing URIs on the Web since they are sparse and not always properly implemented . But ultimately, most linked data is just instance data that can be represented in simpler attribute-value form. By shunning a knowledge representation language (namely, OWL) at the processing end, we have put too much burden on what are really just instance records. Linked data does not get the balance of labor right. It ignores the reality that data consumers want actionable information over being able to click from data item to data item, with overall quality reduced to the lowest common denominator. If a publisher has the interest and capability to publish quality linked data, great! It should become part of the data ingest pool and the data becomes easy to consume. But to insist on linked data across the board creates unnecessary barriers. Linked data growth has not nearly kept pace with broader structured data growth on the Web .
At the enterprise level, the semantic technology stack is hard to grasp and understand for newcomers. RDF and OWL awareness and understanding are nearly nil in companies without prior semantic Web experience, or 99.9% of all companies. This is not a failure of the enterprises; it is the failure of us, the advocates and suppliers. While we (Structured Dynamics) have developed and continue to refine the turnkey Open Semantic Framework stack, and have spent more efforts than most in documenting and explicating its use, the systems are still too complicated. We combine complicated content management systems as user front-ends to a complicated semantic technology stack that needs to be driven by a complicated (to develop) ontology. And we think we are doing some of the best technology transfer around!
Moreover, while these systems are good at integrating concepts and schema, they are virtually silent on the question of actual data integration. It is shocking to say, but the semantic Web has no vocabularies or tools sufficient to enable data items for the same entity from two different datasets to be combined or reconciled . These issues can be solved within the individual enterprise, but again the system breaks when distributed interoperability is the desire. General Web-based inconsistencies, such as in HTML coding or mime types, impose hurdles on distributed interoperability. These are some of the reasons why we see the successes in the context (generally) of single institutions, as opposed to anything that is truly yet Web-wide.
These points, as is often the case with software-oriented technologies, come down to a disappointing state of tooling. Markets drive developer interest, and market share has been disappointing; thus, fewer tools. Tool interest comes from commercial engagements, and not generally grants, the major source of semantic Web funding, particularly in the European Union. Pragmatic tools that solve real problems in user adoption are rarely a sufficient basis for getting a Ph.D.
The weaknesses in tooling extend from basic installation, to configuration, unit and integrated tests, data conversion and lifting, and, especially, all things ontology. Weaknesses in ontology tooling include (critically) mapping, consistency and coherency checking, authoring, managing, version control, re-factoring, optimization, and workflows. All of these issues are solvable; they are standard software challenges. But it is hard to conquer markets largely with the wrong army pursuing the wrong objectives in response to the wrong incentives.
Yet, despite the weaknesses in tooling, we believe we have been fairly effective in transferring technology to our clients. It takes more documentation and more training and, often, accompanying tool development or improvement in the workflow areas critical to the project. But clients need to be told this as well. In these still early stages, successful clients are going to have to expend more staff effort. With reasonable commitment, it is demonstrable that an enterprise can take over and manage a large-scale semantic engagement on its own. Still, for semantic technologies to have greater market penetration, it will be necessary to lower those commitments.How Has the Environment Changed?
Of course, over the period of this history, the environment as a whole has changed markedly. The Web today is almost unrecognizable from the Web of 15 years ago. If one assumes that Web technologies tend to have a five year or so period of turnover, we have gone through at least two to three generations of change on the Web since the initial vision for the semantic Web.
The most systemic changes in this period have been cloud computing and the adoption of the smartphone. These, plus the network of workstations approach to data centers, have radically changed what is desirable in a large-scale, distributed architecture. APIs have become RESTful and database infrastructures have become flatter and more distributed. These architectures and their supporting infrastructure — such as virtual servers, MapReduce variants, and many applications — have in turn opened the door to performant management of large volumes of flat (key-value or graph) data, or big data.
Extremely germane to the semantic Web — indeed, overall, for artificial intelligence — has been the occurrence of knowledge-based AI (KBAI). The marrying of electronic Web knowledge bases — such as Wikipedia or internal ones like the Google search index or its Knowledge Graph — with improvements in machine-learning algorithms is systematically mowing down what used to be called the Grand Challenges of computing. Sensors are also now entering the picture, from our phones to our homes and our cars, that exposes the higher-order requirement for data integration combined with semantics. NLP kits have improved in terms of accuracy and execution speed; many semantic tasks such as tagging or categorizing or questioning already perform at acceptable levels for most projects.
On the tooling side, nearly all building blocks for what needs to be done next are available in open source, with some platform areas quite functional (including OSF, of course). We have also been successful in finding clients that agree to open source the development work we do for them, since they are benefiting from the open source development that went on before them.What Did We Set Out to Achieve?
When Structured Dynamics entered the picture, there were already many tools available and core languages had been released. Our view of the world at that time led us to adopt two priorities for what we thought might be a five year or so plan. We have achieved the objectives we set for ourselves then, though it has taken us a couple of years longer to realize.
One priority was to develop a reference structure for concepts to serve as a “grounding” basis for relating datasets, vocabularies, schema, taxonomies, or ontologies. We achieved this with our first commercial release (v 1.00) of UMBEL in February 2011. Subsequent to that we have progressed to v 1.05. In the coming months we will see two further major updates that have been under active effort for about eight months.
The other priority was to create a turnkey foundation for a semantic enterprise. This, too, has been achieved, with many more releases. The Open Semantic Framework (OSF) is now in version 3.00, backed by a 500-article training documentation and technical wiki. Support tooling now includes automated installation, testing, and data transfer and synchronization.
Because our corporate objectives were largely achieved it was time to look at lessons learned and set new directions. This article, in part, is a result of that process.How Did Our Priorities Evolve Over the Decade?
I thought it would be helpful to use the content of this AI3 blog to track how concerns and priorities changed for me and Structured Dynamics over this history. Since I started my blog quite soon after my entry into the semantic Web, the record of my perspectives was conterminous and rather complete.
The fifty articles below trace my evolution in knowledge and skills, as well as a progression from structured data to the semantic Web. These 50 articles represent about 11% of all articles in my chronological archive; they were selected as being the most germane to the question of evolution of the semantic Web.
After early ramp up, most of the formative discussion below occurred in the early years. Posts have declined most recently as implementation has taken over. Note most of the links below have PDFs available from their main pages.
The early years of this history were concentrated on gathering background information and getting educated. The release of DBpedia in 2007 showed how knowledge bases would become essential to the semantic Web. We also identified that a lack of shared reference concepts was making it difficult to “ground” different semantic Web datasets or schema to one another. Another key theme was the diversity of native data structures on the Web, but also how all of them could be readily represented in RDF.
By 2008 we began to study the logical underpinnings to the semantic Web as we were coming to understand how it should be practiced. We also began studying Web-oriented architectures as key design guidance going forward. These themes continued into 2009, though now informed by clients and applications, which was expanding our understanding of requirements (and, sometimes, shortcomings) in the enterprise marketplace. The importance of an open world approach to the basic open nature of knowledge management was cementing a clarity of the role and fit of semantic solutions in the overall informaton space. The general community shift to linked data was beginning to surface worries.
2010 marked a shift for us to become more of a popularizer of semantic technologies in the enterprise, useful to attract and inform prospects. The central role of ontologies as the guiding structures (either as codified knowledge structures or as instruction sets for the platform) for OSF opened realizations that generic functional software could be designed that can be re-used in most any knowledge domain by simply changing the data and ontologies guiding them. This increased our efforts in ontology tooling and training, now geared more to the knowledge worker. The importance of groundings for aligning schema and data caused us to work hard on UMBEL in 2011 to get it to a commercial release state.
All of these efforts were converging on design thoughts about the nature of information and how it is signified and communicated. The bases of an overall philosophy regarding our work emerged around the teachings of Charles S Peirce and Claude Shannon. Semantics and groundings were clearly essential to convey accurate messages. Simple forms, so long as they are correct, are always preferred over complex ones because message transmittal is more efficient and less subject to losses (inaccuracies). How these structures could be represented in graphs affirmed the structural correctness of the design approach. The now obvious re-awakening of artificial intelligence helps to put the semantic Web in context: a key subpart, but still a subset, of artificial intelligence. The percentage of formative articles directly related over these last couple of years to the semantic Web drops much, as the emphasis continues to shift to tech transfer.What Else Did We Learn?
Not all lessons learned warranted an article on their own. So, we have also reflected on what other lessons we learned over this decade. The overall theme is: Simpler is better.
Distributed data interoperability across the Web is a fundamental weakness. There are no magic tricks to integrate data. Data mapping and integration will always require massaging. Each data integration activity needs its own solution. However, it can greatly be helped with ontologies and with better tooling.
In keeping with the lesson of grounding, a reference ontology for attributes is missing. It is needed as a bridge across disparate datasets describing similar entities or with different attributes for the same entities. It is also a means to reduce the pairwise combinatorial issue of integrating multiple datasets. And, whatever is done in the data integration area, an open world approach will be essential given the nature of knowledge information.
There is good design and best practice for distributed architectures. The larger these installations become, the more important it is to use a lightweight, loosely-coupled design. RESTful Web services and their interfaces are key. Simpler services with fewer functions can be designed to complement one another and increase throughput effectiveness.
Functional programming languages align well with the data and schema in knowledge management functions. Ontologies, as structures, also fit well with functional languages. The ability to create DSLs should continue to improve bringing the knowledge management function directly into the hands of its users, the knowledge workers.
In a broader sense, alluded to above, the semantic Web is but a set of concepts. There are multiple ways to use it. It can be leveraged without requiring “core” semantic Web tools such a triple stores. Solr can act as a semantic store because semantics, NLP and search are naturally married. But, the semantic Web, in turn, needs to become re-embedded in artificial intelligence, now backed by knowledge bases, which are themselves creatures of the semantic Web.
Design needs to move away from linked data or the semantic Web as the goals. The building blocks are there, though perhaps not yet combined or expressed well. The real improvements now to the overall knowledge function will result from knowledge bases, artificial intelligence, and the semantic Web working together. That is the next frontier.
Overall, we perhaps have been in the wrong war for the wrong reasons. Linked data is certainly not an end and mostly appears to represent work, rather than innovation. The semantic Web is no longer the right war, either, because improvements there will not come so much from arguing semantic languages and paradigms. Learning how to master distributed data integration will teach the semantic Web much, and coupling artificial intelligence with knowledge bases will do much to improve the most labor-intensive stumbling blocks in the knowledge management workflow: mappings and transformations. Further, these same bases will extend the reach into analytical and statistical realms.
The semantic Web has always been an infrastructure play to us. On that basis, it will be hard to ever judge market penetration or dominance. So, maybe in terms of a vision from 15 years ago the growth of the semantic Web has been disappointing. But, for Fred and me, we are finally seeing the landscape clearly and in perspective, even if from a viewpoint that may be different from others’. From our vantage point, we are at the exciting cusp of a new, broader synthesis.NOTE: This is Part I of a two-part series. Part II will appear shortly.  Tim Berners-Lee, James Hendler, and Ora Lassila, “The Semantic Web,” in Scientific American 284(5): pp 34-43, 2001. See http://www.scientificamerican.com/article.cfm?articleID=00048144-10D2-1C70-84A9809EC588EF21&catID=2.  For those with a spare 90 minutes or so, you may also want to view this panel session and debate that took place on “An OWL 2 Far?” at ISWC ’08 in Karlsruhe, Germany, on October 28, 2008. The panel was chaired by Peter F. Patel-Schneider (Bell Labs, Alcathor) with the panel members of Stefan Decker (DERI Galway), Michel Dumontier (Carleton University), Tim Finin (University of Maryland) and Ian Horrocks (University of Oxford), with much audience participation. See http://videolectures.net/iswc08_panel_schneider_owl/  Open Biomedical Ontologies (OBO) is an effort to create controlled vocabularies for shared use across different biological and medical domains. As of 2006, OBO formed part of the resources of the U.S. National Center for Biomedical Ontology (NCBO). As of the date of this article, there were 376 ontologies listed on the NCBO’s BioOntology site. Both OBO and BioOntology provide tools and best practices.  Fifth International Workshop on Consuming Linked Data (COLD 2014), co-located with the 13th International Semantic Web Conference (ISWC) in Riva del Garda, Italy, October 19-20.  See http://www.mkbergman.com/category/linked-data/.  See http://www.mkbergman.com/1713/fyn-v-ft/.  See the thread on the W3C semantic web mailing list beginning at http://lists.w3.org/Archives/Public/semantic-web/2014Jul/0129.html.
I have been periodically tracking ontology tools for some time now (also as contained on the Open Semantic Framework wiki). Recent work caused me to update the listing in the ontology matching/mapping/alignment area. Ontology alignment is important once one attempts to integrate across multiple knowledge bases. Steady progress in better performance (precision and recall) has been occurring, though efforts may have plateaued somewhat. Shvaiko and Euzenat have a good report on the state of the art in ontology alignment.
There has been a formalized activity on ontology alignment going back to 2003. This OAEI (Ontology Alignment Evaluation Initiative) has evolved to include formal tests and datasets, and annual evaluations and bake-offs. Over the years, various tools have come and gone, and some have evolved through multiple versions. Some are provided in source or with online demos; others are research efforts with no testable code.
As far as I know, no one has kept a current and comprehensive listing of these tools and their active status (though the Ontology Matching site does have an outdated list). Please accept the listing below as one attempt to redress this gap.
I welcome submissions of new (unlisted) tools, particularly those that are still active and available for download. There are surely gaps in what is listed below. Also, expect some new tools and updated results to be forthcoming from OAEI 2014 as reported at the Ontology Mapping workshop at ISWC effort in October.
Besides the tapering improvement in performance, other notable trends in ontology matching include ways to optimize multiple scoring methods and using background knowledge to help guide alignments.Active, Often with Code
My youngest, Zak, turned 25 today. In med school, being taught by his mom, and in love, he is in a pretty sweet spot. Zak has never known me not working from home. For that matter, Zak’s older sister, Erin, also in med school, is also too young to remember me as anything but being a dad working from home. This week also marks my 25th anniversary of working from home. Pretty amazing!
Over that period of time I have had five companies, lived in four states, have worked either alone or have run companies with up to 35 employees, raised millions in financing, and have mostly continued to move in (I hope) a forward direction. In the early years, I literally commuted between Montana and Washington, DC. Now, I rarely travel. My current company, Structured Dynamics, is pretty simple with low overhead — by design.
In the beginning, it was really not that easy being an independent teleworker. Technology, culture, and practices were very much different than today. In the beginning, I was actually asked to speak to groups about what it was like to be a teleworker from home. The fax machine was the key enabler for us early teleworkers. But the fax was merely the first expression of sending content digitally over phone lines, the same basic model that applies today albeit with many advances. Fortunately, it has been at least ten years since technology was a real challenge in being able to work from home.
Running a company of more than a few individuals (remotely or local) is still perhaps a stupid approach when working from home. The times that I did have remote workforces — true for two of my companies — I tried to compensate by much travel and presence. But, honestly, I don’t think a remote boss can ever really work that well, for either the boss or the employees. Key workers or partners can work well from home, but not someone charged with leading and also setting broader culture in an organization with more than a few employees.Some Notable Events
In 1989 I had been working in downtown Washington, DC, for nearly ten years when I decided to head out on my own. We were getting close to having our second child and we had realized the metro DC area was not the place we wanted to raise our children. My wife, now a biomedical professor, had just embarked on her own career that had limited choices as to institution and location. Working on my own and being my own boss seemed to provide the flexibility that we would need to best meet the needs of two professionals and our family. Though I started my initial energy consulting business in DC right as Zak was born, we shortly thereafter moved to Hamilton, MT, for the second of my wife’s post doctoral positions. We designed our new home with explicit attention to my office and work situation.
The nexus of my original energy consulting was DC (and other national locations), which caused me to travel up to 150,000 air miles annually during the early years. I literally went through a case of thermal fax paper each month to keep current with my clients and partners. In rural Montana I was quite the anomaly; most of my plane telecommuters were celebrities like Huey Lewis or Andie MacDowell, frequent co-travellers on my flights out of Missoula. Local civic groups often asked me to speak on what it was like to be a telework pioneer. I continued this for nearly five years, when I decided that software trumped straight consulting. By the time of our next move to Vermillion, SD, my transition to software was complete.
Vermillion was the locus of my first multi-employee company, VisualMetrics. All of that company’s employees worked from home. Staff meetings were conducted in a converted basement meeting room at my house. Our company sweatshirts celebrated the beginnings in the basement.
That effort was the springboard to starting up another company, BrightPlanet, which extended our employee reach to Sioux Falls, SD. Soon after that founding, my wife was recruited to a bigger medical school at the University of Iowa, which caused us to move to Iowa City, IA, fifteen years ago. We have been here ever since. Both of my kids graduated from the local West High School and then went on to college and their incipient careers.
For half of this period I kept my relationship with BrightPlanet, frequently making the seven-hour drive to Sioux Falls. That was probably the most difficult period of my teleworking tenure, since the company came to have many employees and the drive was pretty exhausting.
By 2006 I returned to consulting, and then was recruited to Zitgist and then formed Structured Dynamics with Fred Giasson. With the start of SD, travel began again. One of the more notable memories of that period was trying to get to a meeting in NYC after the Great Flood of 2008. I earlier wrote about that waterlogged event. I find a pig floating across the Iowa countryside to be an apt image.Some Observations About Telework
Teleworking has both confirmed (and exceeded) certain expectations I had going in, while also seeing some surprises. Some of this I earlier documented in a twenty-year retrospective.
On the expectations front, I knew that I would gain more useful time during the day by avoiding the commute, which was 90 min total for me in DC. What surprised me, however, was the additional time I gained by not needing to keep home and office systems synchronized. Early on, I found I had gained about two hours of time each day in avoiding commuting and coordination activities! Not bad; not bad at all.
My initial experience in DC working from a converted bedroom also convinced me that it was important to have a space separate from the actual living space. Though I also used a converted bedroom in Vermillion, in my other two locations I have designed specific office spaces, somewhat physically removed from the general living space in the home. Two of the other pictures in this piece show my current office in Iowa.
Planning office space in advance means you can tailor the space to your work habits. For me, I want lots of natural light, a view from the windows, and lots of desk and whiteboard space. I also needed room for office equipment (copiers in the early years, fax, printers and the like) and file cabinets. When in Montana, I designed up and had built my own office furniture suite; I have kept that furniture with me ever since.
Teaching myself and the kids that office time and office space were fairly sacrosanct was important, too. Sure, it was helpful to be around for the kids for boo-boos and emergencies and dedicated kid’s time, and to be able to be there for home repairs and the like. But, for the most part, I have tried to treat my office as a separate space and to have the kids do so, too, though this is no longer applicable now that the kids are on their own. While growing up, I think this separation of space became natural to the kids and the family, and my office has always been viewed as mostly separate, though the door has never been closed or locked.
I never eat in my office and do not have a television or other distractions. I try to keep regular hours. I treat my office as such, and keep family and home activities distinct. Through the years the biggest challenge has been to not allow the ease of “going into the office” to take over my life. Frankly, I have not always kept this balance, since I have always loved my work and it is also one of my main hobbies and source of intellectual stimulus.
The biggest surprise over my tenure of working from home has been the Internet and what it has brought to make telework easier. It took a few years from the mid-1990s for this potential to show itself, but it is now so evident as to be unremarkable. All of the advantages that have been brought by cloud computing have caused home work to have nearly the same advantages of a standard work office. Teleworkers can share, collaborate, create, obtain and research equivalently to what a standard office worker may do.
Another pleasant surprise is the decline in meeting demands. Meetings, which are only occasionally essential, require either travel or greater planning in advance in the telework circumstance. Otherwise, it is great to avoid the all-too-frequent office meetings, most of which are tremendous time suckers and black holes for productivity.
Because so much can be done digitally, my general office expenses have continued to decline. Fax and copies are now rarely required. A single Internet line provides email, computing, streaming media, and VOIP benefits. I no longer need to replace my office computer each 2-3 years, and what little office equipment I need (multi-function, etc.) has also dropped greatly in cost.
Another major surprise has been the orders of magnitude reductions in what it takes to start up a new business, especially one based on software. Legal forms and incorporation are now almost commodities. Open source software means new applications and platforms may be assembled much quicker and at much reduced cost. In my early efforts, it was next to impossible to start up a software venture without substantial cash reserves in the bank or some form of external financing, even if just from angels. These efforts used to take tremendous time and effort, and represented a significant overhead cost to actually getting product produced. These impediments have now been largely swept away. (Though external financing still may be useful for marketing and scale-up.)Wondering About Where Work is Heading
Of course, telework is not for everyone. It applies mostly to knowledge work, and where constant interaction or collaboration is not the norm. Individuals who need much social interaction or who are not pretty disciplined or self-starters would find working from home difficult and perhaps unfulfilling.
But the same trends that are making telework easier are also changing the nature of the standard knowledge workforce. One wonders if the heyday of the corporation, made infamous in the 1950s by the gray flannel man, is not an institutional phenomenon in slow decline. Even larger corporations are moving towards mobile workforces and virtual or temporary offices for their knowledge workers (despite Yahoo’s banning of telework).
It used to be (and still somewhat largely is) the case that the attraction of the “city” was in pulling together smart and innovative people in close proximity. Though some of these attractions will never go away, those attractions come at a cost in higher costs of living, congestion, crime and sensory assault. The Internet is enabling virtual and specialized communities to form, some of which due to their niche attractions, may not even be easily sustainable in cities.
We sometimes laugh that we have retreated to an Ozzie and Harriet-like circumstance of safe neighborhoods, great public schools, and a sense of balance and local community. We live in a bucolic spot, with nature and greenery all around us. It may be flyover country, but it is one that has sheltered us from much that has become ugly and challenging in the modern urban environment. Our kids left their hometown for spells of their own to see the big city and touch the elephant. But, they have also gravitated back to a more accommodating and easy pace of living.
I hope that an interconnected world of knowledge will better allow all of us to live in circumstances of our own choosing, ones that help nurture ourselves and our families. For my family, achieving that vision in part has been helped by my being able to work from home. Now, after trying it out for a quarter of a century, I’m convinced I would not have it any other way.
Our first release of the UMBEL site occurred in 2007 while UMBEL was still under development. That site used its own homegrown HTML. The release was followed in 2008 by the addition of our own Web services. The Web services were well-received, which caused Structured Dynamics to develop the more general structWSF Web services framework (most recently updated as the OSF Web Services). We subsequently migrated the earlier UMBEL Web services to this more general framework, and also migrated to Drupal as the standard content management and Web site component for OSF.
For most reasons, including all client work to date, our OSF framework (Web services + Drupal 7) has been performant and met client site needs. However, the operation of the UMBEL Web services was often problematic after moving to the Drupal (full OSF) version. Unfortunately, we have seen both performance and stability problems, though calculations over a full 28,000 node graph are a challenge in any environment.
Since the UMBEL structure was an order of magnitude larger than our client work to date, we have frankly adopted a posture of occasional monitoring and reboots to keep the UMBEL Web site up. This posture was not limiting use of UMBEL for general browsing purposes, but was limiting its usefulness as a working API.
Because the cobbler’s son is often the last to get shoes, we have let the UMBEL Web site chill to a degree in the background. But, now, with other imperatives underway and some dedicated time to look directly at performance of larger-scale ontologies, we have looked at these items anew. The report card on our current evaluations is contained in a newly released UMBEL Web site with services, which I summarize and provide context for below. What emerges is an interesting story of discovery and growth.Basis of the New Site
The new UMBEL site and its underlying 28,000 concept graph is consistent with the OSF layered architecture. However, the Web services are now written in Clojure and the Web site framework uses Bootstrap and plain ol’ HTML. These structured and foundational changes have been championed by Fred Giasson, SD’s chief technology officer, who is also putting forth a blog series on Clojure in particular. He also has a current post from a technical basis on these UMBEL site and service changes.
In essence, we have learned two important things about our prior practice with respect to making UMBEL Web services broadly available. First, for UMBEL, we do not need or want our standard configuration of having a Drupal front-end as the interface into OSF. Access to a knowledge graph does not need — and is ill-served — by having a complicated interface stand atop a large-scale concept model. APIs and Web services are the most important interaction points with the UMBEL knowledge graph, not a user-oriented Web site.
Second, in the various phases of our work, we had come to embrace the idea of ontology-driven applications (what we have termed ODapps). The compelling vision behind such structures is to place the emphasis on knowledge structures and data, rather than more software. Once one begins to unpack that vision, it can also become clear that software programming languages themselves that look toward “code as data” might be one way to be consistent with that vision.Seeking a Sense of Harmony
For years I have been writing about data integration and interoperability and our company has been devoted to the topic. I have written extensively about the importance of RDF and description logics to how we organize and represent data. We were also some of the first to supplement RDF with a faceted text-search engine (Solr) to provide the most responsive query environment across structured to unstructured data. We have also adopted ontologies and the OWL 2 (plus SKOS) languages as standards to both foster and enable interoperability. We have explored native data structs to understand how wild forms of information can be efficiently pipelined into interoperable RDF and text forms.
All of this points to the ideal of the democratization of the information function in the enterprise. In other words, to the idea that how data structures get organized and represented (the ontology side of things) is something that knowledge workers can do themselves, rather than accepting the bottleneck of IT and programmers.
This is well and good except there is a critical “last mile” between data representation and data usefulness. This “last mile” deals with how actual data gets manipulated and then organized and presented (visualized). Query responses, reports, analysis and maps continue to be the choke points between knowledge workers and their IT support. And one need not frame this entirely from an enterprise perspective: these same challenges exist for the individual researcher or the small organization.
So, while one can focus on data and its organization and representation, until we address this “last mile” problem, we still are not likely addressing the largest source of frustration and lost opportunities in the knowledge function.
The reason that simple data struct forms and tools like spreadsheets continue to be popular is that they are empirically the best tools for the “last mile”. Web forms and services are increasingly showing their strengths in this realm.
Once one steps back and looks at the entire cycle from basic datum to actionable knowledge, it is clear that the question of data model is but one portion of the challenge. The remaining challenge is how (now) accessible information can be placed into context and acted upon. Further, if one premise is the democratization of the information function, then the challenge should also be how to provide productive capabilities for the last mile to the knowledge worker. Productivity is enhanced when there are the fewest channels and distortions between signal (problem) and channel (user chains).
Fred, in his investigation of functional languages, clearly saw that bringing the languages of code (programming) into the language of data (knowledge workers as expressed in our RDF world view) was one means to reduce the number and lossiness of the channels between problem (signal) and solution. A world view premised on the efficient representation and interoperability of data must logically support the idea of a coding (instructional or language) base aligned as well to problems. Moreover, since software guides the actual computer operations, a form of the software that supports the nature of the data should also provide a more performant framework for moving forward. In technical terms, this is known as homoiconicity.
Whether one looks to the intellectual foundations of Charles S Peirce or Claude Shannon (both of whom we do), one can see that the idea of signs and information theory means finding both data representations and code that minimize communication losses and promote the accurate transfer of the message. Lossless data transmission is one contributor to that vision, but so too is a functional representation for how the information is to be processed and transformed that aligns most closely with the information at hand.
Ergo, a better model for data is not enough. A better model of how to manipulate that data (that is, software) is also needed that aligns with the idea of coherence and structure in the underlying information. For our purposes, we have chosen Clojure as the functional language basis for these new UMBEL Web services. Not only is it performant, but it aligns well with the creation of domain-specific languages (DSLs) that also promise to democratize the computing function for the knowledge worker.Bringing the Pieces Together
Fred and I founded Structured Dynamics a bit more than five years ago. But, we had worked together much earlier on UMBEL and Zitgist. For nearly ten years now, we have episodically emphasized a few different initiatives and passions.
One of those passions has been the structure of data and information. It is this perspective that brought us to RDF and data structs (and our irON efforts) at various times. The idea of structure is a basis for our company name, and represents the belief that structure can be brought to unstructured forms (via tagging, for example). Structure is perhaps the most common notion or concept in my own writings for a decade.
Another need has been the idea of making semantic technologies operational. We have been keen researchers of the tools space and algorithms and such since the beginning. We observed early on that many innovative and open source semantic programs existed, but most were the result of EU grants or academic efforts elsewhere. Thousands of tools existed, but very few had either been evaluated or stress-tested. By bringing together the best of class tools and integrating them, we could begin to provide a useful semantic platform for enterprises. This motivation was the genesis for the Open Semantic Framework, and has been the major source of our client support since SD was founded. We have finally created an enterprise-capable platform and have done much to transfer its technology. But, these concepts are difficult, and much remains to be done before semantic technologies are a standard option for enterprises.
Still, in another vein, our first love and interest has been knowledge bases. We first identified the need for UMBEL years ago when we perceived an organizing vocabulary would become an essential glue on the Web. We pursued and studied Wikipedia and how it is informing knowledge bases. Instance data and how it is represented is a passion for how these knowledge bases (KBs) get leveraged going forward.
As a smaller consulting and development boutique, we have needed to be opportunistic about when and where we devoted efforts to these pieces. So, over the months or years, we have at various times devoted ourselves to data models and ontologies (structure), the Open Semantic Framework (platform), or UMBEL or Wikipedia (KBs, knowledge bases). Depending on funding and priorities, any one of these threads did receive episodic attention and focus. But, truth is, each one of these pieces has been developed in (project-level) isolation to the whole. Such piecemeal development was essential until each component achieved an appropriate degree of maturity.
I could say we could foresee some years back that all of these pieces would eventually reinforce and bolster one another. Though there is a small bit of truth in that statement, the way things have actually unfolded is to show, as experience and sophistication have been gained, that there is a synergy that comes in the interplay of these various pieces. The goodness is that Structured Dynamics’ efforts (and of its predecessors) were building inexorably to the possible cross-fertilization of these efforts.
Once this kind of realization takes place — that data, code and semantics move hand-in-hand — it then becomes logical to look at the entire knowledge ecosystem. For example, it is not surprising that artificial intelligence, now in the informed guise of KB-backed systems, has again come to the fore. It is also not surprising that what software and programming languages we bring to bear also directly interact with these concerns. Just as Hadoop and non-relational database systems have become prominent, we should also investigate what kind of programming languages and constructs may best fit into this brave new information world.
What we have seen from that investigation is that functional languages (with their DSL offspring) somehow fit into the overall equation moving forward. SD has moved from a single-focus endeavor to one explicitly looking at integration and interoperability issues. What we had earlier seen as (largely) independent pieces we now see as fitting into a broader equation of related emphases:
Structure + Platform + KBs + Functional Language = Knowledge Worker-based Interoperability
We are seeing artificial intelligence moving in these directions. As a subset of AI, I suspect we will also see the semantic Web moving in the same direction.
We clearly now have the theory, the data, the understanding of semantics, and languages and data representations that can make these democratic interoperabilities become real. This new UMBEL Web site is the first expression of how these pieces can begin to work together into a compelling, accessible whole.
We welcome you to visit and to take advantage of UMBEL’s fully accessible APIs.
In my last article on artificial intelligence, I made the statement that “. . . innovation is the source of wealth creation.” I made that unquestioned statement as part of my reflexive world view. But, when I re-read the article after its posting, I asked myself: What are the actual arguments and evidence for this innovation-to-wealth assertion? Surprisingly, there is not nearly the evidential basis for this assertion that I would have assumed.
Since Adam Smith, the signal focus of economics has been its attempt to explain the basis of growth. This is not surprising since the birth of the field of economics also corresponded to an historically unprecedented inflection point in economic growth (see next). Smith ascribed this source to productivity resulting from the division of labor using his famous example of the pin factory. But it is really only within the past fifty years or so that economists have begun unpacking the growth function from the other factors of production.
Growth is a percent increase from a prior state. In economic terms, growth compounded over a period of time has the virtuous reward of resulting in increased wealth. Economic growth is often measured through such means as revenues (for the individual firm) or GDP (for regions or countries). Net worth (for the firm) or GDP per capita or net worth (for individuals) measure the wealth associated with the current stock of economic goods at any given point in time. And, of course, wealth alone also masks the importance of changes in comfort, convenience, freedom, choice, leisure, mobility and other values that may accompany growth and transcend the material. Too, some “externalities” of economic growth may be negative, such as congestion or pollution, but it is also true that wealthier societies tend to regulate against these effects.
Not only have we seen discontinuities in growth (and then wealth) throughout history, but we see them today between individuals, firms, industries, cities, regions and nations. Unpacking the economic factors of production that lead to growth thus has immense importance across the entire economic spectrum — from individuals to nations. Explicating and then managing these factors are intuitively a basis for improving the welfare of any economic actor. Unlocking the nature of growth, or better understanding that nature, should aid in helping to promote still further growth and wealth. Though questions of distribution and fairness may remain, a rising tide lifts all boats.
Thus, understanding the basis of growth, sustained over time, leading to greater wealth for individuals or nations, is the central question facing economics. And, as we see below, that understanding in turn is intimately related to the importance of information and innovation.The Common Sense Argument
If we toil, year by year, doing the same activity, like growing wheat, and we gain the same harvest for the same labor and land and inputs, that is what we expect. Yet sometimes, the weather or rainfall patterns may differ, or we may have more children helping us in the fields, or a mule to help plow. Money helps us buy more of the important inputs, maybe more land, more mules or the comfort to have more children. These are the traditional factors of production: that is, land, capital and labor.
If we add more of these factors to the mix, we still understand we have merely tweaked the standard basis of our wheat production. Differences in the amount of these factors of production, throughout most of human history, are what accounted for the differences between rich and poor, landlord and serf. If, by virtue of having more land or children, we are now able to feed more people, we are by first definitions more wealthy, and if we can accumulate more of this wealth, we can leverage these standard factors even more. When we can keep more of what we produce we become more wealthy. Control and exploitation have been logical paths to much wealth creation.
These factors are pretty easy to observe and track. We intuitively understand that more inputs of labor, land or capital can themselves result in growth, but a growth that feels and appears rather fixed based on the change in these inputs. This kind of growth has a more-or-less trending return based on changes in these inputs. These types of inputs may also be subject to diminishing returns, wherein adding more of a given factor produces diminishing or negative payoff. For example, adding more fertilizer to the wheat crop produces less per unit output yield after some optimum, and then can actually reduce yields by burning the crop. Or, while a computer increases the productivity of an individual worker, giving her more computers may actually degrade her overall performance.
But there is also clearly a different kind of growth that is not constrained to a fixed or declining return based on inputs. Perhaps we have a neighbor that raises more wheat, possibly on drier, more marginal land, or with less water or fertilizer. His yield exceeds our own. These differences occur because our neighbor is doing something different and is producing more given his inputs.
Innovation is an individual affair in its discovery, but a communal one in its application. (At which point it is known as information.) Better ways of planting or spacing the wheat, perhaps using a plow, or selecting certain wheat strains for next year’s plantings, or irrigating the land, or providing harnesses to the mules, or dividing and specializing the responsibilities amongst the children, can result in real differences in how much gets harvested for a “similar” set of inputs. And, what I initially innovate, becomes information for the next farmer to emulate. Some of these innovations are new devices, such as harnesses or plows. Some of these innovations are new practices, such as tilling or irrigation methods or specializations in tasks or labor. And, of course, not every farmer must innovate on his own. Copying and imitation diffuse these changes across farms and workers.
Truly, for millenia, this is how human progress took place. Some innovations, such as fire, the wheel, iron and bronze, the arch, alphabets, the plow and the yoke had material benefits to all who encountered them. These innovations were fundamental and diffused at the pace of human movement. But, one could argue, each was understood to be a flash of insight, and not a product of systemic information and process. Further, innovations tended to diffuse slowly, along the pace and concentration of trade routes. The innovative event was quite rare, and most practices had been stable for centuries. It is not at all surprising that early economic ideas tended to focus on the traditional factors of production of land, labor and capital. These had been the steady constants for what had been very slow growth for centuries.
But then a real discontinuity in economic growth compared to all previous recorded history occurred in the early 1800s. Historically flat income averages skyrocketed, as this famous figure showing global changes in per capita (person) GDP from Angus Maddison illustrates .
William Nordhaus has captured a similar discontinuity looking at the price of light, normalized according to the labor effort needed to obtain 1000 lumens of light. It, too, shows an exponential decrease in the price of lighting beginning about 1800 :
These comparatively abrupt changes in growth rates, and concomitant changes in wealth, that were orders of magnitude higher than what had been experienced before in human history, garnered the attention of economists and economic historians as never before .
From the beginning, this difference in growth rates was largely attributed to “technological change”, but the specific causes of this change have been ascribed to many things. The close concurrency to the Enlightenment suggested some fundamental change in thinking. Similarly, the concurrence with the Industrial Revolution suggested the importance of machines, prime movers and the harnessing of energy. Cultural and religious factors have been posited to explain why Britain and then the United States were the initial centers of growth. The invisible hand of the market and division of labor and specialization were advocated by Adam Smith. I have argued the importance of the mechanical printing press and pulp paper in bringing information to a broader swathe of society . Education and support for basic and applied research have their advocates. Financial and banking innovations, and the rule of law and patents and other intellectual property rights, have also been cited as causes.
Common sense tells us that all of these factors, and perhaps more, can all work as force multipliers to the traditional inputs to the economic function.
But, until the mid-1950s, the broad sense of “technological change” and vague causative factors were more often than not argued in an anecdotal, literary way. Empirical datasets were few and far between to test hypotheses, and quantitative means of reasoning over economic problems were only just beginning. Economic growth theory was only just beginning to be an economic discipline in its own right.
Joseph Schumpeter, in The Theory of Economic Development, first published in 1911, argued that innovation was central to economic growth and constantly disrupted the general equilibrium of market exchange . Innovation granted the firm a temporary monopoly status in which to charge higher rents, thereby providing an incentive for further innovation. Schumpeter’s emphasis on entrepreneurialship and his popularization of “creative destruction” recognized that new innovative market entrants may cause older firms to become obsolete. He tied these ideas into his basic views on business cycles, also driven by technological change. Innovation was central to Schumpeter’s economic world view.
But the theoretical story really begins in earnest after World War II when the hidden X factor of technological change — in what came to be expressed as total factor productivity — came to the fore to complete the economic growth equation .The Exogenous Model
Robert Solow is an American economist particularly known for his work on the theory of economic growth; the exogenous growth model is named after his work. Solow took courses from Schumpeter at Harvard and was influenced by his views on innovation and technological change , though Solow was also part of the generation of economists embracing the new discipline of mathematical or quantitative economics, which was foreign to Schumpeter.
As noted, economic growth was known to go beyond the typical factors of production. Solow’s insight in two papers in 1956 and 1957, for which he won a Nobel prize, was that technological change, what he called “technological progress,” must be the “residual” left over from empirical growth once the traditional inputs of labor and capital are removed . Using his model, Solow calculated that about 87.5% of the growth in US output per worker was attributable to technical progress . A substitute term is total-factor productivity (TFP), the “residual” in total output not credited to the traditional inputs of labor and capital. By definition, TFP cannot be measured directly.
We can express this mathematically as showing total output (Y) as a function of total-factor productivity (A), capital input (K), labor input (L), and the two inputs’ respective shares of output (α and β are the capital input share of contribution for K and L respectively):
These considerations make the exogenous growth model one of the neo-classical growth models, wherein the long-run rate of growth is exogenously supplied, apart from the internal growths of labor and capital. Within this camp, one explanation is based on the savings rate (the Harrod–Domar model); the other, as shown herein, is the rate of technological progress (Solow-Swan model ). By definition, in either of these so-called neo-classical models, the savings rate or the rate of technological progress remains unexplained. They are abstract external forces that are just “out there.”
The TFP approach remains strong as a basis for estimating total non-traditional inputs to the production function. It also provides a specific target within quantitative economics to begin addressing explicitly a placeholder for innovation, technological change, information, or other non-traditional considerations for what constitutes the overall production function. But, frankly, TFP still is a blob that needs to be unpacked and teased apart.The Endogenous Model
A seminal paper by Kenneth Arrow in 1962 paper introduced the concept and evidence for what he called “learning by doing“; what is now more formally understood and accepted as the learning curve. Unlike a specific innovation, the idea of the learning curve captured that experience and practice led to efficiencies and productivity. In other words, more whatever could be done with less what as we learn better how to do whatever.
By the 1960s and 1970s it was becoming clear that developed economies were becoming information economies, increasingly staffed by knowledge workers, and these forces needed to be made explicit within quantitative models. Robert Lucas, now a Nobel laureate from the University of Chicago, probed the questions of rational expectations and internal factors promoting growth. By the mid-1980s, a group of growth theorists had became increasingly dissatisfied with common accounts of exogenous factors determining long-run growth. The focus shifted to the needs for quantitative models that made these “technological” or “information” factors explicit. In other words, these “X” factors needed to be moved from a lump, external consideration to an internal one within the models, with their own multipliers and feedbacks. In short, these new growth factors needed to be made endogenous (internal), not exogenous (external).
A book by David Warsh in 2007, Knowledge and the Wealth of Nations: A Story of Economic Discovery, is a comprehensive explanation of this transition, with a focus on Paul Romer, then of Stanford University, but earlier a colleague of Lucas, pivoting on his seminal paper, “Endogenous Technological Change” . By bringing the consideration internal to the model, it could be groped, inspected and broken into parts.
Besides this essential change in focus, this and related Romer papers also brought two further key insights. First, information and its artifacts are also products and outputs of the economic function. And, second, once produced, many information or knowledge assets may be produced or distributed at essentially zero marginal cost. A new dimension in “rival” and “non-rival” goods had been added to the growth theory lexicon. Information and knowledge themselves were becoming both inputs and outputs to the economic function. This understanding required still further unpacking.Refining Inputs and Parameters
As a non-economist, it seems a bit perplexing to me how long it took the discipline to start explicating and unpacking the factors of economic growth . To be fair, most every domain of human inquiry has suffered from lacking essential test datasets and statistics upon which to probe and test assumptions. There is perhaps no better poster child for this lack of reference datasets than what has been necessary to test and probe the questions related to economic growth. Yet, as our intro suggested, there is also perhaps no more important area of human inquiry than to understand these non-traditional factors of economic growth. Better understanding of these factors will impact all economic actors from individuals to firms to nations.
Our first approximation must be to get to common units and denominators that enable calculation and comparison. Things like GDP, for example, need to be re-expressed as per capita figures to take out general population growth; money terms need to be expressed in real dollars (or whatever currency), perhaps even further adjusted to account for differences in assumed deflators and inflators across metrics. We’re getting smart enough about this stuff that we can now apply best practices for common data comparisons.
Even the traditional factors of production need further attention. Let’s first take the concept of labor. Labor is ubiquitous in virtually all economic calculations.
Most economic datasets compare items across space and time. A simple labor adjustment to per capita or hours worked can mask these underlying structural changes: life expectancy of the workers; male-female participation in the workforce; hours worked per week; holidays and vacation time; changes in retirement ages; general population and cohort growth; and, then and only then, labor productivity. Of course, the reasons for labor productivity itself come back to innovation and information: the use of better machines, practices and methods by which we do our tasks.
Similarly, the idea of “human capital” has also become predominate in the economic growth literature. Is human capital a subset of general capital? labor? Does human capital include education, training, experience, intellectual capabilitiies, etc.? And, if so, how can these be measured and made consistent for comparison or decision purposes?
We also see that the nature of innovation, information, knowledge, intellectual property (IP), practices, information artifacts, and the like, lack any consistency as to definitions and boundaries. How can nebulous concepts be compared to still other nebulous concepts in order to draw meaningful conclusons? How can test datasets be created to refine these questions if the basic concepts and definitions remain ambiguous?
We see, for example, that knowledge and its role in economic growth may vary as to whether the knowledge is propositional (the ‘sciences’), prescriptive (‘recipes’), a discovery, or an invention . These may not be the best splits, but clearly we must be able to distinguish at minimum innovative ‘aha!s’ from the tech transfer of best practices. These are fundamentally different notions of information. And, of course, none of this discussion directly addresses the internal controversy within the economics community of information v knowledge.
Once we normalize our traditional inputs to the economic function to appropriate per unit bases expressed in constant, real dollars, the residual “total factor productivity” is all due to innovation and information. Innovation is the spark that brings us new methods and devices for doing things, as eventually disseminated throughout the economy via the diffusion of information. Since innovation is itself based on information, we can truly say that information is the fount from which all per capita growth and wealth ultimately derives.The Empirical Argument
In a recent paper on total factor productivity going back 150 years to the Civil War, researchers from the Congressional Budget Office have calculated that private-sector nonfarm TFP in the United States grew at an average rate of roughly 1.6 percent to 1.8 percent annually, but has experienced several surges occurring in varying parts of the economy .
On a different basis, I have used Robert Schiller’s published data on per capita GDP going back to 1900 to show a similar growth trend . The trendline from this data series shows an annual compounded growth rate of about 1.84% per year:
These kinds of growth rates imply a doubling of wealth every 40 to 45 years.
When TFP was first being formulated, Solow calculated that 87.5% of the growth in US output per worker was attributable to technical progress . In 1954, Solomon Fabricant estimated 90% of growth was due to technological factors . But, as we have seen, these were “lumpy” measures and factors like the changing size and composition of the work force (especially the growth of women and two-earner families) also masked other changes.
A different way to approximate the role of technological progress is to look at the market measure of the US markets. Again using Schiller’s CAPE data , but also now adjusted to a per capita basis (for the US, ), we see the following trends since 1900:
Nominally, labor is removed from this equation because it has been accounted for as an expense on the firm’s books. Similarly, the return due to capital has been accounted for via the payout of dividends. Under these bases, we see that the growth in value of large US firms — despite the severe oscillations due to market cycles — has been a bit more than 1 per cent per annum compounded. This would suggest that the combination of innovation and information accounted for about 55 percent of the overall per capital GDP growth rate noted earlier.
But this proxy is itself flawed in many ways. First, the S&P index is for only the 500 largest US firms, which are certainly not representative. Also, comparing GDP and S&P figures hides the fact that much of the growth and productivity of US firms occurs via foreign subsidiaries. Also, of course, labor and capital productivity — themselves the result of innovation and information — are also taken out of the S&P estimates. The discrepancy between TFP estimates as a source of growth and intrinsic S&P valuation growth is in part explained by this different accounting metric. But the real issue in all of these proxies is that we are not yet fully unpacking the various sources of information and innovation as the drivers of underlying growth.
Only within the last few years have we begun to assemble the right datasets and account for the right factors in this unpacking of growth factors. For example, between 2000 and 2005, estimates at the industry level indicate that almost half of aggregate productivity was due to productivity growth originating from information technology , though the IT industries themselves only accounted for a little over 3% of nominal aggregate value .
These findings are from a more detailed analysis of productivity and growth by Jorgenson, Ho and Samuels . Their analysis attempted to explicitly separate out innovation from the diffusion of prior innovations due to information. In the authors’ words:
“We show that the great preponderance of economic growth in the US since 1947 involves the replication of existing technologies through investment in equipment, structures, and software and expansion of the labor force. Contrary to the well-known views of Robert Solow (1957) and Simon Kuznets (1971), innovation accounts for only about twenty percent of US economic growth. This is the most important empirical finding from the recent research on productivity measurement surveyed by Jorgenson (2009). “
I think some of these differences are due to semantics and terminology. Remember, early residuals and TFP discussions were centered around the concept of “technological progress”. What Schumpeter referred to as “innovation” is now understood to be too broad; innovation is but a part of the overall growth effect due to information. What is helpful from these more recent studies is to separate out innovation from information dissemination. The next step, for which we have not yet developed useful datasets, would be to unpack the ideas of innovation and information into the categories from Mokyr . Namely, these are discoveries and inventions (innovation) and the ideas of propositional and prescriptive information first distinguished by Michael Polanyi as tacit knowledge.
The aphorism that we can not understand what we can not measure applies here. To take our understanding of these empirical factors to the next level we will need to refine our concepts and gather defensible data for estimating them. A proper accounting for growth should also likely distinguish transformative innovations (such as the printing press, electricity and computing) from other discoveries and inventions.The Beautiful Synergy of Innovation and Information
By 2009, Romer and Jones were able to claim that the endogenous growth model had been proven, and they put forward six research questions to look for in the coming 25 years, including the role of human capital, differential growth rates between countries, and accelerated growth . Innovation had finally assumed its central, internal role in understanding growth.
Innovation is the root source of new devices, new technologies, new practices, new methods and new theories. Innovation, in turn, is based upon the foundational substrate of information. As new innovations occur, new information is added to this substrate, all in a virtuous circle.
Markets will rise and fall, and business cycles will gyrate. New businesses and business models will emerge while others are destroyed or whither away. These reflections of animal spirits and uneven (imperfect or wrong) information can never be smoothed. But, the trajectory of growth, fueled by the beautiful synergy of innovation and information, points to an optimistic future.
To be sure, I am not positing a near-term upward trend in the stock markets. In fact, my own personal view is that markets are temporarily oversold, with a higher near-term probability of declines rather than rises. These oscillations are part and parcel of market cycles. My longer-term optimism reflects more fundamental trends.
We are all aware of the explosion of information and content. Today, like the broadening base of information and literacy that I have elsewhere posited as a major factor in the first upward inflection of economic growth in the 1800s , we are in the midst of a still newer — and optimistic — inflection point. Digital content and the Internet are bringing information to nearly every human on earth. Assistive technologies are bringing this information to those previously shut out due to disabilities in sight, hearing or mobility. Non-rivalrous goods can be duplicated at essentially zero cost and open source and broad access mean new ventures can be assembled and tested in the marketplace with unprecedented speed at unprecendented lower cost. Innovation is no longer the remit solely of an educated elite, but is available to every thinking person on earth.
These are all harbingers of continued growth and increases in wealth. Sure, ignorance, despotism, fanaticism and prejudice will cause some periods and pockets to be shut off from these trends, but the broad sweep of information and history looks assured.
Innovation, as Schumpeter first posited a century ago, grants the firm a temporary monopolistic advantage. In a time of openness, information growth, and universal access to that information, the winning competitive formula for firms and knowledge workers alike is constant innovation. Though a commitment to innovation leads to a bumpy path, it is an upward one, and most assuredly the path that is on the right side of history. The historical data were originally developed in three books by Angus Maddison: Monitoring the World Economy 1820-1992, OECD, Paris 1995; The World Economy: A Millennial Perspective, OECD Development Centre, Paris 2001; and The World Economy: Historical Statistics, OECD Development Centre, Paris 2003. All these contain detailed source notes. Figures for 1820 onwards are annual, wherever possible. For earlier years, benchmark figures are shown for 1 AD, 1000 AD, 1500, 1600 and 1700. These figures have been updated to 2003 and may be downloaded by spreadsheet from the Groningen Growth and Development Centre (GGDC), a research group of economists and economic historians at the Economics Department of the University of Groningen, headed by Maddison before his passing in 2010. See http://www.ggdc.net/.  William D. Nordhaus, 1996. “Do Real-Output and Real-Wage Measures Capture Reality? The History of Lighting Suggests Not,” in Timothy F. Bresnahan and Robert J. Gordon, eds., The Economics of New Goods, University of Chicago Press, ISBN: 0-226-07415-3, January 1996, pp. 27 – 70. See http://www.nber.org/chapters/c6064.  I have addressed these broad topics firstly in, “Information is the Basis for Economic Growth” (Adaptive Information blog, AI3, August 23, 2007), and in some book reviews, notably “Knowledge: Unravelling the X Factor in Growth and Wealth” (Adaptive Information blog, AI3, June 21, 2006) and “Historical Origins of the Knowledge Economy” (Adaptive Information blog, AI3, July 6, 2006).  William Lazonick, 2013. “The Theory of Innovative Enterprise: A Foundation of Economic Analysis,” AIR Working Paper, #13-05/01, 36 pp., May 2013. See http://www.theairnet.org/files/research/WorkingPapers/Lazonick_InnovativeEnterprise_AIR-WP13.0501.pdf.  “The growth of growth theory,” from The Economist, May 18th 2006.  “Robert Solow on Joseph Schumpeter,” in Economist’s View, Thursday, May 17, 2007. Retrieved on June 11, 2014.  Solow’s endogenous model of economic growth is also known as the Solow-Swan neo-classical growth model as the model was independently discovered by Trevor W. Swan and published in “The Economic Record” in 1956, allows the determinants of economic growth to be separated out into increases in inputs (labor and capital) and technical progress.  Robert M. Solow, 1957. “Technical Change and the Aggregate Production Function”. Review of Economics and Statistics (The MIT Press) 39 (3): 312–320. doi:10.2307/1926047. JSTOR 1926047.  Published in the Journal of Political Economy in 1990.  In 2002 Joel Mokyr, an economic historian from Northwestern University, wrote a book that should be read by anyone interested in knowledge and its role in economic growth. The Gifts of Athena : Historical Origins of the Knowledge Economy is a sweeping and comprehensive account of the period from 1760 (in what Mokyr calls the “Industrial Enlightenment”) through the Industrial Revolution beginning roughly in 1820 and then continuing through the end of the 19th century.  Robert Shackleton, 2013. “Total Factor Productivity Growth in Historical Perspective,” Working Paper Series, Congressional Budget Office, 21 pp., March 2013. See htttp://www.cbo.gov/sites/default/files/cbofiles/attachments/44002_TFP_Growth_03-18-2013.pdf.  Stock market and cyclically-adjusted price earnings (CAPE) ratio data from Robert J. Schiller, 2000. Irrational Exuberance, Princeton University Press. Data as periodically updated and available from http://www.econ.yale.edu/~shiller/data/ie_data.xls.  Solomon fabricant, 1954. “Economic Progress and Economic Change, a part of the 34th annual report of the National Bureau of Economic Research.” New York.  CAPE per capita adjustment from http://www.multpl.com/united-states-population/table?f=m.  Steven Rosenthal, Matthew Russell, Jon D. Samuels, Erich H. Strassner, and Lisa Usher, 2014. “Integrated Industry – Level Production Account for the United States: Intellectual Property Products and the 2007 NAICS,” May 15, 2014 (preliminary), 24 pp. See http://scholar.harvard.edu/files/jorgenson/files/jorgenson_ho_samuels_worldklems_2014_0519.pdf.  Dale W. Jorgenson, Mun S. Ho, and Jon D. Samuels, 2014. “Long-term Estimates of U.S. Productivity and Growth,” pepared for Presentation at Third World KLEMS Conference Growth and Stagnation in the World Economy, Tokyo, May 19-20, 2014. See http://www.worldklems.net/conferences/worldklems2014/worldklems2014_Ho.pdf. Charles I. Jones and Paul M. Romer, 2009. “The New Kaldor Facts: Ideas, Institutions, Population, and Human Capital,” Working Paper 15094, National Bureau Of Economic Research, 31 pp, June 2009. See http://www.nber.org/papers/w15094.
When I inaugurated this AI3 blog in 2005 I made this statement in the about section to clarify that the “three AIs” stood for adaptive information, adaptive innovation, and adaptive infrastructure, and not the AI of artificial intelligence:
. . . I personally believe artificial intelligence to be a lot of hooey and hype at best, and a misnomer and misdirection at worst. . . . ‘Artificial intelligence’ is a misdirection of attention and energy.
Gulp. OK. Time to take my medicine.
I am today formally retracting those statements — probably should have done so some time ago — and want to explain why. As much as anything, it has to do with the changing understanding of what is artificial intelligence, recently affirmed by global-scale applications and technologies, working effectively right now.Many Winters within AI
Though the idea of automatons and intelligent agents standing in for humans is about as old as human storytelling, the real basic ideas around artificial intelligence became current as part of the World War II effort and were finally given a name in a famous 1956 conference at Dartmouth. Initially namers and advocates of artificial intelligence included such founders as John McCarthy, Herbert Simon, Claude Shannon and Marvin Minsky. Money to support early interest in artificial intelligence came from the part of the US military that eventually became ARPA (now DARPA), with the funding going to individual researchers to use as they wished as opposed to specific projects. Along with many futuristic visions of the 1950s to 1970s, the promises for artificial intelligence were bold, including being able to capture and automate most notable basic human capabilities.
Popular movies and books promoted the ideas of autonomous robots that we could speak with and command and that would anticipate our needs and wishes so as to act as simulacrum agents lessening our burdens and adding to our leisure and capabilities . Algorithms would be discovered and codified that would mimic the basis of human thought and intelligence. The idea of the Turing machine established a defensible basis for foreseeing that any problem of mathematical logic could be captured and taken on by computers.
The predictable failure of this vision to deliver caused a backlash, sufficient that the US Congress prohibited further open-ended funding via the Mansfield Amendments in 1969 and 1973, such that by 1974 AI funding in the US had largely dried up. Similar restrictions were applied to the British research community. The result of this backlash caused the first of what would prove to be many “winters” of funding and acceptance for AI.
Roughly a decade later, in response to the perceived Japanese threat for “fifth-generation” computing in the mid-1980s, a number of AI programs were again funded. While hardware developments were proceeding apace, efforts around McCarthy’s AI-oriented language Lisp and common sense logic frameworks (what are now called ontologies or knowledge graphs) such as Cyc began to receive sponsorship again. The mid-1990s were the time of “expert systems,” to be populated by knowledge engineers charged with interviewing internal subject matter experts (SMEs) to codify their knowledge for later reuse. These efforts, too, disappointed in terms of the lack of practical benefits delivered. More AI winters ensued.
AI (“artificial intelligence”) came to again lose its credibility. Some researchers moved into specific algorithmic disciplines — Bayesian statistics and neural networks predominant — while others shifted into such areas a “hyperlinks” and what became the semantic Web. Today, one could argue, that the lost mojo of AI has affected those in the semantic Web in almost a dialectic way. First, there are those who embrace the idea of intelligent agents and global knowledge structures, more-or-less in keeping with some sort of vision of artificial intelligence. Second, there are those that have seen the failures of the past, do not want to repeat them, and are more inclined to support “loosely bounded” structure focused on bottoms-up assertions. OWL modelers and ontologists tend to occupy the first camp; linked data advocates more the second camp.
The natural community for knowledge representation and management has thus tended to bifurcate a bit: global, “visionary” AI types, with history to overcome and challenged by the sheer scale of what emerged from the Internet; and incrementalists, happy to accept a bit of RDF structured data in the hopes of an ongoing evolution to more structure and interoperability.
Ten years ago, when I made the conscious decision to reject the AI of artificial intelligence as a label for this blog, an algorithmic-vision of AI seemed “wrong” and not in keeping with the general trends of the Web. That was the basis and justification for my then-statements on AI. But a funny thing happened on the way to a cogent forecast: a massive disruption called the Internet came about that — while it took a decade to gestate — changed the whole underlying substrate over which AI could take place. Like so much of history, innovation had presented to us an entirely different reality upon which to “understand” and develop artificial intelligence. It is those changes — plus the fruits from them — that is defining AI in a new light.Eight AI Megatrends
There are, by my reckoning, at least eight major trends that have been improving AI’s prospects, especially over the past decade (Numbers #3 to #7 below are quite related to AI, the other three are general trends.) Some of the proven wonders we now see in use such as speech recognition, speech synthesis, language translations, entity recognition, image and facial recognition, computer vision, question answering, autocompletion and spell correction, recommendation systems, sentiment analysis, information extraction, document categorization, natural language processing, machine learning, reasoning, optical character recognition, word sense disambiguation, search and information retrieval, and text generation and summarization, with their many additional categories and sub-categories, are proof these trends are making a difference. None individually constitutes what may be called “AI”, but, in combination, they show compellingly that much of AI’s initial vision is indeed being fulfilled to some degree and in some specific aspect today.
Nearly all of these applications correspond to the Grand Challenges for symbolic computing identified in the 1980s. Until a decade ago, very few of them save search and initial NLP were producing results with sufficient quality and accuracy. Now, all are.
In the past ten years, most evident in the past five, tremendous breakthroughs have occurred across the entire spectrum of artificial intelligence applications. We can point to at least the eight following megatrends enabling these breakthroughs.#1 Computer Power
A constant river of innovation has fueled the logarithmic power improvements in computers since the first transistor. Moore’s law has led to massive improvements in hardware cost, numbers of computation cycles, and amounts of bits stored. Networking capabilities are now truly global and numbers of interconnected devices exceeds billions. Computer software innovations lead to faster and better procedures and methods; as a category, software innovation likely exceeds hardware improvements as a source of computing productivity. What today fits in the palm of our hand thirty years ago required entire rooms, and did not do one billionth of what can be done today.
The rich savanna of computing has itself encouraged a bloom of innovations, many of which contribute to artificial intelligence prospects.#2 The Internet (and Web)
Though clearly a related function to the general improvements in computing and hardware, the advent of the Internet and its more relevant offspring of the Web has had, I believe, the most fundamental impact on the change in prospects for artificial intelligence. The sheer scale of the Web network has made available crowdsourced innovations like Wikipedia and other crowdsourced data and knowledge bases. More broadly, global content across the entire Web, accessible via a common HTTP protocol, multiplied every individual’s access to information — pay close attention — by a factor of a billion or more.
Because the entire Web is interconnected, the sheer raw grist of connected data available to analyze such things as relatedness or similarity is gamechanging. Manual constructs and derived relations from years past can now be multiplied and magnified at Web scale. Any relationship test or validation can be accomplished nearly instantaneously and at (essentially) zero cost. Phenomenal!#3 Expectations
The discrediting of AI and its holdover smell has itself been a factor working in its favor. By being discredited, it has been possible for multiple possible AI components, many listed herein, to be developed and attended to in relative isolation. Each of today’s current pieceparts to AI could be focused upon on their own, without taint from the broader “AI” brush. Because the constituents were recognizable and justifiable on their own, they did not need to fulfill the past overblown visions and expectations for “AI” writ large. The pieceparts could develop in peace.
This observation, if true, means that grand visions like “artificial intelligence” are perhaps rarely (ever?) the result of a grand top-down plan. Rather, like a good stew, it is individual components that need to mature and become available to create the final meal. Since these ingredients need to stand or contribute on their own for their own purposes, the actual resulting stew may vary as to its ultimate ingredients. If one ingredient is not ripe or available, we vary our recipe according to what is available. There is no one single recipe leading to a tasty stew.
Put another way, AI has been flying under the radar for at least the last ten to fifteen years. Portions of the older AI agenda have benefited from specific attention. Better still, the new emergence of the idea of artificial intelligence is also more toned down and practical. Artificial intelligence is now, I believe, understood to be part of a process and not some autonomous embodiment. Human interaction and communication are themselves imprecise and subject to error. Why should not be artificial means to boost those same human capabilities?.
From the standpoint of expectations, artificial intelligence has evolved from science fiction to essentially zero awareness, meanwhile delivering, on a broad scale, focused wonder capabilities such as (nearly) instantaneous translations across 60 leading human languages.#4 Global Knowledge Bases
How can a system promise useful suggestions or alternatives if it is bereft of information?
At the local or personal level we well understand that we need to describe ourselves via attributes, the more the merrier in terms of a more complete description. A pretty good record for me would include such things as physical description, image, work and economic description, family and life description, education description, text narratives from fun to historical, etc. The more complete description of me requires many sources and many attributes and many perspectives. But, of course, I do not live alone in the world. To describe my world, which constantly changes, I need to describe other thousands of entities I encounter daily. Each of these, too, has many attributes and relationships to other entities. Each of these entities also changes over time (has histories) and place. So, context becomes another critical dimension.
The growth of the Web at scale has resulted in some tremendous knowledge bases of entities and concepts. Freebase and Wikipedia are two of the best known, but virtually every domain has its own sources and richness. These knowledge bases, in turn, are often open for use by others. Text mining and digital data mean these data can be combined and made to interoperate. That process is only just beginning.
Though early efforts in artificial intelligence understood that capturing and modeling common sense was both an essential and surprisingly difficult task — the impetus, for example, behind the thirty-year effort of the Cyc knowledge base – what is new in today’s circumstance is how these massive knowledge bases can inform and guide symbolic computing. The literally thousands of research papers regarding use of Wikipedia data alone  shows how these massive knowledge bases are providing base knowledge around which AI algorithms can work.
The abiding impression is that the availability of these data sources has fundamentally changed how AI is done. Unlike the early years of mostly algorithms and rules, AI has now evolved to explicitly embrace Web-scale content and data and the statistics that may be derived from global corpora.#5 Deep Learning
Machine learning is a core AI concept used to determine discriminative characteristics or patterns within source input data. It has been a constant emphasis of AI since the beginning.
Various machine learning algorithms — such as Markov chains, neural networks, conditional random fields, Bayesian statistics, and many other options — can be characterized among many dimensions. Some are supervised, meaning they need to be trained against a standard corpus in order to estimate parameters; others require little or no training, but may be less accurate as a result. Some are statistically based; others are based on pattern matching of various forms.
A more recent trend has been to combine multiple techniques in what is known as deep learning, where the problem set is modeled as a layered hierarchy of distributed representations, with each layer using (often) neural network techniques for unsupervised learning, followed by supervised feedback (often termed “back-propagation”) to fine-tune parameters. While computationally slower than other techniques, this approach has the advantage of automating the supervised learning phase and is proving generally most effective across a range of AI applications.
More fundamentally, there is a virtuous circle of feedback occurring between AI machine learning algorithms and reference knowledge and statistical bases (see next). This can extend the accuracy, completeness and efficiency of supervised methods. Some notable academic departments have relied on Web-scale corpora (University of Washington and Carnegie Mellon University are two prominent examples in the US). The most dominant player in this realm, however, has been Google (though all of the major search engine and social networking companies have smaller initiatives of similar character).#6 Big Statistical Data
Using both statistical techniques and results from machine learning, massive datasets of entities, relationships and facts are being extracted from the Web. Some of these efforts, such as the academic NELL (CMU) or KnowItAll or Open IE (UWash) involve extractions from the open Web. Others, such as the terabyte (TB) n-gram listings from Google, are derived from Web-scale pages or Google books. These examples are but a sampling of various datasets and corpora available.
These various statistical datasets may be used directly for research on their own, or may contribute to further bootstrapping of still further-refined AI techniques. Similar datasets are aiding advertising placements, search term disambiguation and machine (language) translation. In some cases, while the full datasets may not be available, open APIs may be available for areas such as entity identification or tabular data.
What is important about these trends is that data, statistics and algorithms are all now being combined in various ways with the aim of achieving acceptable AI-backed results at Web scale. It is really via the combination of these techniques that we are seeing the most impressive AI results.#7 Big Structure
A more nascent area, really in just its first stages of effectiveness, is the application of “big structure” to all of this information. By “big structure” I mean the application of domain and knowledge graphs to help arrange and place the concepts and entities at hand.
At Web scale, the early Yahoo! directory and Open Directory were the first examples of structuring domains. Wikipedia next became the most widely used category structure; Freebase, for example, used Wikipedia to initially bootstrap its own structure. A portion of Freebase is now what is used for Google’s own Knowledge Graph. DBpedia also created its own ontology out of the infobox structure of Wikipedia. The major search engines have also put forward the schema.org structure as a means of (mostly) organizing entity and attribute information and structured data. schema.org putatively is an input to the Google Knowledge Graph, but the exact mechanism and ability to trace the results is pretty opaque.
The need for big structure is rapidly emerging as one of the key challenges for Web-scale AI. The Web and crowdsourcing appears well suited to being able to generate entity and attribute data. What remains unclear is how this information can be coherently organized at the scale of the Web. This problem is becoming acute, because the success of “big data” on the Web needs to ultimately find an organized, coherent expression in the aggregate. This is one major AI challenge that remains distinctly unsolved, though promising first steps exist.#8 Open Source and Content
The major theme of these AI breakthroughs comes from leveraging the global content of the Web. And this enabler, in turn, has been critically dependent on the open source nature of AI algorithms, software code and code infrastructure and architecture, and open content and (generally) open APIs. Open code, algorithms, datasets and knowledge have expanded the pool of human intelligence that can be brought to bear on the question of artificial intelligence. The positive feedbacks greased through open channels of information, code and data have been absolutely essential to the amazing AI progress of the past few years.
To be sure, open does not mean a level playing field. (See discussion on Google, next.) But, without open source and open content and data, I think no one could argue that progress would have been anywhere near as rapid as it has been. The synergy arising from open source and content has thus been another essential factor in the recent and rapid progress in AI.The Race to Intelligence
Since innovation is the source of wealth creation, it is also no surprise that the megatrends surrounding AI have also drawn significant investment interest. This interest is in the form of a race to acquire the most innovative AI startups and human expertise (capital) in AI. Since Google has been my common touchstone in this piece — and because Google is the biggest gorilla in the room — we can use them to illustrate the scope and pace of this race. (Though Amazon, Facebook, Microsoft and IBM are also clearly entrants in this race.)
A number of recent articles, notably ones in the Washington Post and The Economist, have highlighted the total dollars at stake in this AI race. Over the past few years, there have been perhaps more than $20 billion in AI-related company acquisitions, with Nest Technologies (Google, $3.2 B), Kiva Systems (Amazon, $775 M), and DeepMind (Google, $660 M) some of the largest.
Within Google alone, there has been a buying spree in search improvements (~ $1.4 B total), robotics ($80 M), machine synthesis and recognition ($250 M), machine learning ($700 M), smart devices ($3.6 B), compression technologies ($200 M), natural language processing ($80 M), and a smattering of others ($50 M), not to mention its internal efforts in self-driving cars. I don’t monitor Google on a constant basis and likely missed some major and relevant acquisitions, but it does appear that Google has perhaps spent over $6 billion over the past five years or so for AI-related acquisitions .
As important as start-up acquisitions has been Google’s commitment to hire and partner with many of the leading AI researchers in the world. Besides the strong partnerships Google maintains with such institutions such as the University of Washington, Carnegie Mellon University, MIT, Stanford, UC Berkeley and others, it has also staffed its research ranks with prominent names from those institutions and others.
Peter Norvig, one of the early advocates for combining algorithmic and statistical AI, joined Google in 2001 and is now its Director of Research. Most recently and notably, Ray Kurzweil joined Google as Director of Engineering in 2012. Other notable AI researchers at Google include Alon Halevy (FusionTables), Ramanathan Guha (schema.org), Geoffrey Hinton (deep learning), Evgeniy Gabrilovich (search and machine learning), and many others for whom I am not as familiar with their research. There is probably more AI talent combined at Google than has ever been assembled in one institution before.
With IBM’s Watson getting its own division and Facebook funding an AI center to the tune of $10 B, plus Apple making a similar commitment to robotic manufacturing, it is clear that all of the major players in the computing space are making big bets on AI moving into the future.AI is Itself But One Beneficiary of These Trends
Since the early winters in artificial intelligence, a phenomenon has developed called the “AI effect“. It really has meant two different things.
First, AI researchers have tended to call their research anything but artificial intelligence. One of the broader and trendy substitutes is known as cognitive computing. Many of the domains and disciplines I noted above got their names and prominent use as substitutes for what used to be labeled as AI. In any case, we can see that AI indeed is a big tent with many components and thrusts.
Second, the “AI effect” also refers to the fact that once an AI technique is embedded in some everyday use, it is no longer perceived as something AI and is taken as a given. Douglas Hofstadter expressed the AI effect concisely by quoting Tesler‘s Theorem: “AI is whatever hasn’t been done yet.”
I was perhaps right to initially reject the algorithmic-centric view of AI from the early years. But now, when matched with big data, big statistics and big structure, all embedded into phenomenal advances in computing power, it is also clear that we are dawning into a new age of AI. One only needs to look at the wondrous progress on many of what had seemed to be impossible Grand Challenges over the past five years to gain an appreciation of the pace and breadth of new developments to come.
These developments will reify and foster similar emphases in semantic technologies, graph structures and analysis, and functional programming and homoiconicity (“data as code, code as data”) that my colleague, Fred Giasson, is now actively exploring. We will find that representational paradigms and the basis of how our tools and algorithms work will increasingly align. There appear to be natural underpinnings to these phenomena, including the pivot of language and meaning, that are closely aligned with the thoughts and writings of that great American pragmatist and logician, Charles S. Peirce. We will increasingly come to see that the wondrous innovations of self-driving cars, talking smartphones, warehouses of fulfillment robots, and computer vision systems can trace their roots back to basic truths of how to see and understand our world.
Understanding these forces will, themselves, help to formulate guidelines and ideas that can foster further innovation. So, in the end, while I still don’t like the term of “artificial” intelligence, it is merely a sign or a term. Adaptive innovations expressed by machines are simply part of the intelligence and structure embodied in the universe, for which we are now gaining the tools and understanding to exploit. Douglas Adams‘ Hyperland is a great exposition on this vision, with my 2007 blog post pointing to the online video.  Wikipedia maintains its own page of research that relies on Wikipedia; I have earlier captured about 250 selected sources called SWEETpedia that relate specifically to semantic technologies and AI.  These are merely estimates, and likely quite wrong in many specifics. The estimates were compiled by reviewing a listing of Google acquisitions (since 2009), supplemented by individual company searches when the acquisition amounts were not listed, followed by analysis of Google’s SEC Edgar filings in a manner similar to this analysis (which was also used for the robotics estimate).