Did We Somehow Misplace Our Knowledge?
Technologists provide no clear path to use what we know when we need it, a particular governance challenge
With Miroslaw Manicki
Humans. We are an odd lot, to be sure, but we embody certain salient characteristics. We learn. Accordingly, we adapt. Associated with those, we communicate (Számadó and Szathmáry, 2012). We all do these things (Brown, 1991).
Furthermore, humans develop tools to support these efforts. We even make tools to make tools — something that is unique in the animal kingdom (Brown, 135).
Recent developments among such tools are computational machines. These have proven useful in certain obvious ways, but poor at others. To use an analogy, rather than supporting “ready, aim, fire”, they result in “ready, fire, aim”. They are notoriously imprecise, especially where complexity and changing contexts conspire against them.
We might wonder with the current onslaught of promotional efforts in support of artificial means of thinking and acting, we might be tempted to wonder if human thought stopped recently and no one noticed.
According to certain technologists, we have lost our edge in cognition and its uses. This can be seen based on the concept of knowledge extraction as described below.
Knowledge extraction is the creation of knowledge from structured (relational databases, XML) and unstructured (text, documents, images) sources. The resulting knowledge needs to be in a machine-readable and machine-interpretable format and must represent knowledge in a manner that facilitates inferencing.
Although it is methodically similar to information extraction (NLP) and ETL (data warehouse), the main criterion is that the extraction result goes beyond the creation of structured information or the transformation into a relational schema. It requires either the reuse of existing formal knowledge (reusing identifiers or ontologies) or the generation of a schema based on the source data (Wikipedia, July 11, 2023).
According to the knowledge extraction, substantive human knowledge exists, but it needs to be gleaned and cobbled together from a variety of sources before it can be used. Also, to be sure, the knowledge extraction task is also referred to as “creation of knowledge” which would indicate that more is happening in the knowledge extraction process than simply rearranging materials for presentation purposes.
This is a heady assertion. There are many dynamics and requirements in identifying relationships in nature and society and coming to understanding their prospects in reality. It is important to refer to knowledge creation as it is considered, documented, and carried out. It is questionable, to be sure — an effort to resolve a presumed problem that would have better solutions if indeed true.
As seen in the next two images, there are both qualitative and quantitative aspects to the effort. These demonstrate limitations of machine-only approaches.
Qualitative research is more effective in questions of relevance and context; quantitative research helps with respect to precision. As seen in the next image, qualitative research involves softer information forms, where quantitative data is typically numerical.
Here we see means by which machines, it is hoped, might jumpstart the processes of knowledge creation. As presented, it is a heady prospect. Can computers make qualitative judgments reliably? How could they, when many of these would involve preferences and tastes and other intrinsic aspects of being human?
The knowledge extraction does apparently refer to existing sources of knowledge, but it still uses creative act in the defining knowledge. This is a vague distinction at best; how can existing data be considered useful when human conclusions are obviously considered inadequate? How can such ‘decisions’ be evaluated from the framework of existing qualitative and quantitative methods? From a quantitative standpoint, how can internal and external validity be relied upon, along with other sticky and difficult design and interpretive judgments (Campbell and Stanley, 1963)?
The DARPA ASKEM project is central to the AI concept and its various implementation strategies in that it is a near-term project, and it focuses on this central question of knowledge extraction. There are four steps to the program:
ASKEM will develop and demonstrate technologies in the following four technical areas (TAs):
TA1: Machine-assisted knowledge discovery and curation
TA2: Machine-assisted modeling
TA3: Machine-assisted simulators
TA4: Workbench for HMI and Integration (I2O, 2021, 7).
Think of it: TA1, the first step involves “knowledge discovery and curation”. This raises a question: If knowledge needs to be discovered, how was it lost in the first place?
Similarly, there is a question of where knowledge might be found. Is it under a rock? Is it on a dusty shelf? Is it on someone’s messy desk?
ASKEM promoters clarify the question: The knowledge to be discovered is in documents and in databases. As can be seen in the following image of a stack of documents, they are a poor means of conveying knowledge in any timely matter. To be understood, they must be read and correlated in some fashion. Lacking an easy way to do this in the current information technology environment, the task will not be carried out (Davis, 1989).
The call came out early in the history of computing for such tools (Sorter, 1969), but it has yet to be answered (Tingey, 2014).
What about the people who are ‘wandering around’ in the general area of knowledge? Might they know something about the knowledge in question?
Of course, they do. They have methods and methodologies, along with longstanding commitments to knowledge creation and maintenance. ASKEM documentation makes reference to such people passively, and not in salient terms. Sadly, human efforts are ignored in the telling, robbing experts as a class from the fundamental elements of expertise. This is a particular weakness of the ASKEM proposition and that of AI in general.
Here is a description of the capabilities of experts:
Experts excel mainly in their own domains.
Experts perceive large meaningful patterns in their domains.
Experts are fast; they are faster than novices at performing the skills of their domain, and they quickly solve problems with little error.
Experts have superior short-term and long-term memory.
Experts see and represent a problem in their domain at a deeper (more principled) level than novices; novices tend to represent a problem at a superficial level.
Experts spend a great deal of time analyzing a problem qualitatively.
Experts have strong self-monitoring skills (Chi, Glaser, and Marshall, 1988, xv-xx).
How can the proponents of knowledge extraction in particular, and AI in general, consider experts in such a condescending fashion and with such disdain? Of course, they make such charges with few consequences, largely due to the economic clout of the technology sector and the fear, uncertainty, and doubt long established with regard to computer design.
If such assertions are accurate — as demonstrated in the following account — the human race amounts to an immense collective of knowledge-starved lemmings whose prospects hang in the balance. The implication is that humans created tools in the form of computers that will need to take over now, that we collectively can no longer engage in such tasks. Somehow, by claiming that computers have more capacity than before, they imply that humans have less.
The use of scientific and expert modeling of all sorts fails frequently, and the failures can have real consequences. The current process of knowledge discovery, model creation, and simulation is highly human-intensive. Notionally, subject matter experts (SMEs) comb through existing knowledge, typically in the form of scientific publications or partial model implementations in code, to discover useful artifacts for their work. These artifacts are studied and the knowledge is manually extracted to inform the SME’s modeling process. Once a computational model is developed, SMEs conduct simulation experiments with the model for validation or prediction.
The results inform a SME’s understanding of the system and drive them to seek out new sources of knowledge to fill gaps and update their models. This highly idealized iterative process is rarely how things proceed in the real world, where the process is recursive and has complex dependencies and biases at every step (e.g., limitations in a SME’s domain knowledge, modeling, or software engineering experience can constrain or bias their actions in other steps).
Errors, omissions, or inherent human biases often go undetected and can be compounded by the process of iteration. This expert knowledge pipeline is slow or broken at every stage, from the black-box simulators used to support analysis, through the semantically-opaque models from which they are built, to the rapidly changing knowledge used to develop and maintain these models. Models go out of date, become hard to maintain, are poorly understood, and are difficult or impossible to evaluate for fitness-for-purpose. The implications of these failures are felt throughout our scientific and technological development and decision-making (I2O, 2021, 4).
Of course, such an important set of statements should be accompanied by copious and persuasive citations — supporting not only the assertions thus made, but also the conclusions. In this case, however, there are none. They are holding the entire human knowledge process in question and doing so blithely with no citations or external evidence of support. This is hubris and malpractice of the highest degree.
This is not a new development. I document an example of this I had in conjunction with a DARPA project conference regarding “rapid knowledge formation” in San Diego in 2003 (Tingey, 2023). The point then and now is that there is a failure to consider the question from the perspective of experts and authorities in all fields. It isn’t that experts are capable or not, it is that AI proponents failed to get experts to participate in there programs. Rather than fix the problem, they decided to ignore them and attempt to ‘pick their minds’ by scanning and reconstructing their material.
Rather than develop the tools needed to resolve the problems in question, they are thus bluffing. My sense is that they are constantly hedging their bets knowing that they will ultimately fail. They hope that true solutions do not otherwise surface; they know that in an environment of fear, uncertainly, and doubt as they promote, they can respond to failure by quietly laying back until there is another opportunity to make their case, such ultimate failure once most everyone has forgotten.
In the absence of a real set of tools and resources to resolve this process, they can repeat this cycle often. There has already been something like a dozen efforts on their part to force widespread adoption of AI.
Furthermore, the phrase “highly human-intensive” in the prior quotation seems intended to convey a negative connotation by the authors. In a sense, the knowledge problem is couched as a ‘messiness’ issue. Humans from this perspective are simply incapable of combining and extending their knowledge. Humans in this sense as a whole are seen as weak and ineffectual by some within their ranks without any documented effort to support their assertions.
What they should be saying here, as referenced earlier in this essay, is that computing tools are inadequate to the task. They aid humans in creating complex models with some nuance and tremendous algorithmic computation power and instantaneous reach, but with limited capacity as currently constituted to support our growing cognitive endeavors.
If there are limitations in the use of knowledge by humans, wouldn’t it be beneficial to address them directly, certainly before working to jury-rig a means of circumventing the human cognitive process? The idea that humans are universally feckless and impotent in this regard is troublesome. If it is a sign of feelings of inferiority on the part of the authors, it is certainly not representative of a general resignation of the point by society generally, nor by persons and institutions specifically charged with knowledge acquisition and dissemination.
The 2022 DARPA ASKEM program refers to none of these. How can knowledge creation be under way when the process is itself held into question? We don’t need computers to create knowledge, we need them to help us to hang onto what we know and make use of it while we constantly improve on that store of knowledge.
References
Brown, D. E. 1991. Human universals. New York: McGraw-Hill.
Bogdan, R. C., and Biklen, S. K. 1982.1998. Qualitative research for education: An introduction to theory and methods. Boston: Allyn & Bacon.
Campbell, D. T., and Stanley, J. C. 1963/1966. Experimental and quasi-experimental designs for research. Boston: Houghton Mifflin Company.
Chi, M. T. H., Glaser, R., and Marshall, J. F. 1988. The nature of expertise. Hillsdale, New Jersey: Lawrence Erlbaum Associates, Publishers.
Davis, F. D. 1989. Perceived usefulness, perceived ease of use, and user acceptance of information technology. MIS Quarterly, 13(3), 319–340, doi:10.2307/249008.
Information Innovation Office (I2O). December 2, 2021. Broad agency announcement automating scientific knowledge extraction and modeling (ASKEM). Washington, DC: U. S. Defense Advanced Research Projects Agency.
Sorter, G. H. 1969, January. An “events” approach to basic accounting theory. The Accounting Review, 44(1), 12–19.
Számadó, S., and Szathmáry, E. 2012. Chapter 14: Evolutionary biological foundations of the origin of language: The co-evolution of language and brain. In M. Tallerman and K. R. Gibson (Eds.), 2012, The Oxford handbook of language evolution. Oxford, UK: Oxford University Press, 157–167.
Tingey, K. B. 2014. The solution: Permanescence. Logan UT: Profundities LLC.
Tingey, K. B., and Manicki, M. 2023, April 20. Experts tend to not be stupid. Medium. https://medium.com/@ken-tingey/experts-tend-to-not-be-stupid-81be64977394
Wikipedia contributors. 2023 April 15, 04:42 UTC [cited 2023 Jul 11]. Knowledge extraction [Internet]. Wikipedia, The Free Encyclopedia. Available from: https://en.wikipedia.org/w/index.php?title=Knowledge_extraction&oldid=1149901938.