4th — Machine learning and data

Emergence/surge and decisions  

By: John M. Switlik, March 2024; Context: Sperry Univac Knowledge Systems Center.

Foreword – Interim remark

This series has one intent: document KBE’s (knowledge-based engineering’s) role, in part, with respect to a major project that will mostly be circumstantially mentioned. A recent academic article went through the history of AI and concluded that there was no AI winter in the 1970s. A related article mentioned the AI winter was seen in the late 1980s and early 1990s.

Looking at KBE, there was no winter. In fact, from the late 1980s until the 2000s, there was increasingly demand for the benefits of KBE. These were real and demonstrable. Too, the work of KBE helped initiate discussions about truth engineering. Who can look around now and not have some notion that things are afloat in ways unseen until now?

One theme that will be prevalent is that the KBE work was industrial with goals of accomplishing whatever the program needed. Too, KBE emphasized support for humans who were making the decisions. As such, KBE needs attention. It never went away. KBE is embedded in some systems. Too, many current themes acknowledge that they came out of or were motivated by KBE.

There is no anti-academic aura about the viewpoint being expressed nor in the interpretations of the data that will support the coming discussions. Rather, there is the caution for two things: care in handling complicated themes, especially those supported by mathematically-driven algorithms; knowledge is a multi-disciplinary affair that branches more than academic departments.

In short, academia does not own truth. Who does? It’s on the table for discussion.

Foreword – recap

The first three of this series were instructional as it required thinking about the many aspects of AI over about seven decades, with five of personal experience. (Note 1) One theme dealt with supporting capable and educated humans in their work which increasingly required computational support. During a large timeframe, this involved working in knowledge-based engineering (KBE) which allowed me to observe engineering closely in practice on a major project: Boeing 777. Prior to that, my career line went from being a user of software related to the work of applied statistics and operations research to increasing involvement in algorithms, a lot of which dealt with program interfaces. A theme that developed over time was to have systems devoted particularly to data management.

As well, there was a transfer of a work focus on decision analysis for management support to a more technical role in support of advanced computing. Along the whole of the work experience, software was a major concern, including development maturity, coding, and extended support. A major transition began with the availability of workstations.

I had the fortune to work for a few decades with Lisp systems as a major resource. By then, local facilities for communications had advanced sufficiently to allow the development of APIs beyond those related to data management. One might say, distributed was in the air which can be noted by SUN’s use of this phrase (paraphrase): the network is the computer. In KBE, Lisp, on the workstation, provided the modeling capability which had external access so that the user’s perspective was enhanced by data management systems as well as major algorithmic support via engineering and mathematical libraries. 

That is the framework behind the viewpoint being expressed in this series. Now, another aspect of the experiential aspects was that the work culminated in discussions related to truth which will be discussed further. Looking ahead, we will argue the point that truth has to be engineered, especially with the advent of the computer and its support for communications. The motivation for the series directly ties to the events of the last few years that put more emphasis on the machine than on the users and their processes.

Partly, much of this comes from the advent and subsequent unfolding of the internet and its facilitating of the world-wide web. With the three decades of experience now under our belt, is it not time to step back and do a thorough review? This would need to be more than science and engineering in action. Cultural shifts are at hand, as well.

The initial announcement of this series was that AI is not solely ML. Essentially, what we saw was that improving the tool set (mental) for the engineer resulted in remarkable accomplishments. Given the industrial setting, there never has been adequate discussion of those experiences. To bring lessons to view, the next pieces of the series looked at some issues that would be important.

The next part was titled “Knowledge and truth” for several reasons. The overview theme of the work had been knowledge. ML brings in a focus on the machine’s roles and on the various concerns of capturing knowledge. Our intent had been to use a human-in-the-loop to optimize the computational work. There are many reasons why this was a choice. With the new approaches seen last year, the topic of emphasizing the need for the human user (in the loop) has risen on the awareness stack.

One thing about using the computer has been the need to handle models that are built for the digital world. In lots of places, models on the computer have become quite complicated. One might even use sophisticated. In engineering, models that include the ability to handle data from the world via sensors plus the facility to simulate using the data have shown remarkable results. We mentioned the self-driving car. There are many examples in aviation. The phenomenon has brought about new concepts, one of which is the “digital twin” which we mentioned.

One might note that of late, some see the “digital twin” as more real than that which it is twinning. Okay, this brings up various discussions. On the one side is caution of not running too fast down a dangerous road. With engineering, one can test the situational aspects quite well. On the other hand, culturally-depicted twins can be problematic for several reasons.

Additionally, using engineering as the best example, this role is one of the professions that is regulated. Engineers need to pass through academic and internship regimens that impart knowledge and test the acquisition. As an aside, we strive for similar routines for the “automated” entities that are now extant and will be coming to fore. Robotics will bring lots of the issue to awareness.

So, the theme of human support is a strong one, depending upon the aspects involved. But, KBE, definitely, took the proper role. As such, this piece of the series touched briefly upon KBE today. Explicit search for “KBE” seemed to have mostly brought hits from outside of the US. But, on looking at other approaches, one found lots of these disciplines recognizing the influence on them from the KBE work. Multi-physics is a common theme now. It involves an effort to improve the modeling of physics with respect to the details that the computer is trying to model.

In terms of gaming, one can observe lots of improvements in how characters are represented from the early stickmen to the current attempts at mimicking natural movements plus taking advantage of advances of graphical processing. 

The next piece of the series looked at physicalness and mathematics, briefly. The first one deals with that which is twinned. Digital twins have no physicalness. They might be tied to algorithms that handle input and cause reactions. That type of thing was the basis for control theory which predated ML. We will have to discuss this further. But, the range of discussion will be large, bordering on philosophy on the one side and the limits of our knowledge of physics (and what it represents) on the other.

A common theme will be mathematics which is core to ML which is the focus, initially. At some point, we will have to branch out to a more balanced type of framework. In the meantime, though, there needs to be attention given to issues that relate to mathematics and its ways. There is a school of thought that needs to be lifted to awareness. Namely, the constructive mode that has become important to use of computing models. On the other hand, there has been tremendous amounts of work using metaphors. One of these is infinity, though computing deals with the finite. Needless to say, we must bring into the discussion how decisions are impacted by the framework in use. At the same time, advanced techniques in complex systems use imaginary numbers which mostly, but not always, shuffle out to allow only the real aspects to remain relevant in a decision situation.

Some of this is indirect and not visible to stakeholders. Our intent is to argue that the mathematics needs to be entirely visible, albeit we need to learn to transform the exposition to a condition that is more amenable to attempts at explanation. Though this is a requirement known to many, there needs to be continuing review and analysis associated with this topic in more than truth engineering modes.

Recent events have shown some GenAI incidents where the behavior was not what we expect of an adult. So, watching a GenAI descend to the mode of a child flopping around on the floor is not awe inspiring. Rather, it reinforces the notion that the lack of control was immaturity in action. It’s like, the acorn doesn’t fall far from the tree. We can bring the necessary taming influence. Yes, the bad behavior (threats) was accompanied by graphics that seemed to awe some. It is a good time to compare this type of fantasy with the real requirement of engineering. So, let us proceed so that we can get to work on improvements.

Continuation of the series – #4

Machine learning, emergence/surge

One thing that became apparent last year was that machine learning (ML) is an old discipline. I dove into it long ago. Reading Minsky, of course, was an influence. I usually want to make up my own opinion about something. So, as we will see with the next pieces of the series where knowledge comes back into focus, my work with KBE allowed some dabbling with ML. But, KBE’s emergence was in the context of higher-order modeling within engineering which is a superior discipline (to be discussed) in applied science. Not being an engineer, my focus was support where it was natural to go toward the details of modeling as seen by humans and the computer systems that they are using. We’ll get back to that. KBE was of the 1970s and 1980s. Before then, computing was too immature for true interactive approaches as we see now. Rather, it crunched using approximates such as FORTRAN allowed.

In short though, evolutionary approaches, I found, had levels of thinking that were not available to those who want such. ML’s emergence after a decade of fiddling with new methods brought by gaming somewhat caught me by surprise. So, I had to go back to refresh my take on the matters, albeit there had been a whole lot of mathematical work done on my part over the past two decades. My reactions were from afar and from reading of their main approach. Plus, Minsky went to his grave without giving them his blessing. Not that such was needed.

ML? It goes back, even those interested in cybernetics thought along this line. It was those probing biology who led the way to the current state. Gaming? They forced the improvements in modeling geometry and differential actions based upon the model. Again, I saw that in the 1980s and 1990s in the context of differences twixt a model-based approach such as KBE and that of the simulation approach based upon numeric processes. Turns out that I would get involved in moderating representational issues to keep schedule on a large project as a major goal, but that gets ahead of the current topic.

The story of ML is old and told elsewhere. The surge? Again, by the time that I saw it, I had been aware of the improvements. I don’t game. To me, truth engineering was the key issue as recent events suggest. In any case, data science was there with its problems. Wrangling? Yes, ex post facto cleanup might be appropriate. And, analytic pushes toward pretty graphics. We watched that evolve over the decades since the start of SIGGRAPH (Note 2) Again, a picture might be worth a thousand words. But, Einstein’s little equation? Try to picture that, simply.

That was an aside? No, scale came to fore with the internet. That was the global view of some minds who were unbalanced from the start. ML’s emergence and surge will give us an opportunity to relook at that. Blame misuse of mathematics by computer science and others. That’s on our track.

Hint or reminder. The first parts of this series are merely to mention some of the important themes involved with grappling with issues that need attention. No one seems to have offered this in the manner that is important to sustainability. Of course, blame it on Covid and brain fog, perhaps. But that work since the 2010 timeframe (more or less) was way before Covid came about. More apropos to the discussion is Silicon Valley’s gradual decline since IP was let loose; then, our associated diminishing all around can be seen as hugely correlated.

After this piece, we will get back to AI as a whole in which ML plays a part. But, people are going to be the key issue. With knowledge of huge importance.

Now, we ought to go into how ML works or attempts to in relation to humans and their cognitive modes. However, it’s been covered many places. This series is taking another approach that is founded upon real work involved with engineering. And, as we said, that discipline is the one who does changes that impact us all; that is, if we are talking a sustainable economy and future. Do people have a say? Yes. We will get there later.

However ML or its partners in crime want to model us, we humans do not know as much about ourselves as we would like to think. But, the BoBs know even less. Now, humans are many and varied and live across a wide world. One of the wishes of ML wannabes is to use data that has been captured with respect to all (or a whole lot) of these ones over the past two decades plus of the internet.

Backing up a little, the ML that we are discussing in particular is the xNN/LLM approach that has come to be known as GenAI. Its appearance took the world by storm. And, there was immediate attention given. Massively. Too, several alternatives were given to the people, most with free access. So, lots of people had the experience; oodles wrote of their experience. It wasn’t long before some issues rose.

This year brought further ado as the notion was pushed further. November of 2022 was the entry. I became aware in February of 2023. By December, there were all sorts of things going on. This piece is jumping over those, for now, as some discussion is going to be necessary. Trust went out the window in 2023. Now, people say, don’t believe an image until you can prove it. Getting a grip on the provenance is one huge step.

One manager of fame said in December of 2023, I’m jumping into AI in a big way in January of 2024. It has lots of potential. Oh yeah. Now, as of about six weeks ago, he is warning people about this stuff being hard. In actuality, it is a step back. But, that is smart, though he could have looked last year. Perhaps, he had someone do the analysis and was misled.

This series is setting the stage for discussion and decision support with regard to how ML ought to go into the future. So, back to the technical look. ML represents a particular approach to implementing certain aspects of mathematics on the computer in a large scale. As some said last year, that is fine for entertainment but not for knowledge. The first person to express that idea in public wrote that was in December of 2022. So, kudos to the guy and to the publisher.

But we are not scoffing about the artificial means to intelligence. ML went awry. I could have told them back in the ‘90s how to go with this stuff. KBE was the example then. It has been on the shelf since then for the most part. So, KBE can help get ML back into the game of life and pursuit of happiness.  

KBE was the milieu in which truth engineering emerged. This was the 1999/2000 timeframe. Since then, KBE became a shadow of its former self. Now, it’s a geometry tutor in the design space. Not knocking that. Geometry has always been my forte since I took solid geometry in high school with the teacher who a Ph.D. in mathematics. After this piece of the series, I will go over a patent (Note 4) that depicts the problem and some notion of continuance. Too, though, a major project of KBE will be looked at briefly which will include mention of ML approaches that were accomplished.

A subject being overlooked so far is psychology the discussion of which would bring several questions to the table. Naturally, one might want to consider what the intelligence is that is to be replicated by the artificial. This has been and will be a continual topic since so much is involved with several sides arguing their point. Do we know our intelligence and its origins? Depends upon the viewpoint, but science has an interest here. On the other hand, though, much of the research with respect to implementations to date has been commercially oriented which can be problematic.

For starters we can assume that conditions being right, the computer will mimic what we see as intelligent behavior. What are those conditions? And how do we measure? Too, given that we can establish the framework, can we give meaning to what we discover? It is easy to recognize what might be termed as hype. As well, one has to acknowledge the affinity of the human mind, on the part of many persons, for the artificial. At the same time, there have been incidents where a human claimed that GenAI was conscious, mainly due to their responses to its outputs. Mostly, these situations involve interactive modes that have become more demonstrative in ways that might suggest that a “creature” is involved. There is a lot to the subject. So far, we see no integrative scheme that would try to establish a sane approach and rational viewpoint. That is, beyond the usual games related to academic posturing. However, a business approach has been the key driver for this research which brings up other questions needing some type of attention.

At some point soon, we will get to the necessary dialog one step of which would be identifying topics to work with. But, what is the core piece of ML or any science? That is the next topic on the table which is approached from the viewpoint of KBE’s role as being representative.  

Data and decisions

Again, this deals with huge subjects. It was complicated even before the advent of the computer and its clouds and services. What has happened with the evolution of storage that is persistent, though, is an accumulation of stuff that relates to about all aspects of human life including their activities. Especially do we see lots of scientific data at hand; then, there are data of a medical variety; business has created its share which it took to managing for analysis after 2000 went by; and so forth.

Then, there are books and articles in periodicals of all sorts. Quite frankly, we all got used to the continual advent of sources. Many were free for access. Then, commerce took hold of the situation so that islands became the norm abetted by clouds with their proprietary cloaking.

All that represents the storage and retrieval of things in digital form. Taking the last, there are many ways to represent something on the computer. This goes along with choices to improve the means for algorithms to do their work. People did these types of transforms in the past.

Taking KBE, the context, as will be seen, was designing a new plane using computer-added systems (CAD, CAE, etc.). Prior to that, some work had been computable, but most of it was manual. Going back, it was all manual though automation of calculation has been going on for over 100 years. IBM just turned 100. They dealt with calculators of a sophisticated nature. For engineers, the slide rule dealt as the personal computer.

At the same time, encoding methods changed. In fact, in the beginning there were all sorts of decisions made that impeded general data science. But who thought that such was important? Airlines, perhaps. The evolution of the reservation system is an example (Note 3). Military? Yes, look at the records of the U.S. Army, for instance, going back to the U.S. Revolution.

Well, engineering had this stricture from the beginning. So, that was a push to KBE as by the 1980s, computer databases had matured to be effectively used in large projects. Engineering is heavily mathematical which helped with the computing aspects somewhat. Complexity and complications were natural to such large affairs.

And, data can be of types such as that we have that which allows millions of youngsters to play games whether locally or globally. That is recent and mentioned as the impetus for better graphics came from the game industry and its users. But, engineering’s data for graphics is different. We will get into that.

One reason to mention the types is that the core of computing has no real foundation. Basically, it has been an ad hoc affair even with the efforts of CompSci. Yet, for the most part, the evolution of computing has been positive. This series is being typed from a desk with a laptop tied to the communication system which is a humongous thing. One might say monolith. But, the only thing monolithic about the computer has been introduced by decisions of humans.

Yet, there are reasons for this that we will get into. We must to do that in order to discuss the importance of truth engineering and the maturing of computing. As we look now, the silos of computing may be thought of as homogenous by those maintaining such. And, what is a silo and how many are such remains to be handled. The issue is there and needs attention.

Computing, everywhere, needs to handle heterogeneously-framed decisions. What does that mean? For short, nothing is easy and canonical. Decisions are hard to do even in easy situations. But the computer exacerbates all of these issues.

To look at the matter, let us bow to mathematicians. This field and its importance was mentioned in the last piece of the series. The context of interest now is that about all of the magic of math comes from it handling what is called homogenous situations. And, if that condition is not there, it has to be achieved.

Some might scoff at my characterization as being too negative. That is part of the discussion. My career was in data and its uses. My KBE work was with data. In numeric routines, representational states are not simply defined. There are situations that are not of concern. Business processes are an example; finance is such; operational processes go back to science. And, engineering deals with it all.

Our problem? The world is heterogenous; our desire is for it to be homogenous. Who will win that battle?

That is only one issue. Assume that we can normalize and get homogenous. Even with that, there is the fact that knowledge itself has been partitioned severely. We can, and will, use the departments of universities/colleges for discussing the issues. Say, an airplane? How many sciences and types of engineers are involved in pulling together such a thing so that it can fly people for many years with constant performance day by day?

ML dropped the ball by trying to cover knowledge with its immature model that assumes too much homogeneity among other things. On the other hand, we see some types of delivery despite the core issues. Why is that? We will see why as we go through the remainder of this series and decide what need to be looked at.

Too, knowledgeable people need ML. Again, the KBE experience can show that. We will do this analysis and discussion using a framework that is decades old. And, we’ll do this by taking it off of the shelf, dusting it off, and then bring modern methods to bear on it.

Will it stand up? Yes indeed.

Coming topics, again

We are setting the stage to discuss associative relationships that seem to have gotten lost in the computing age. Next up, we will look at a patent (U.S.) that represents some of the important issues of modeling and truth engineering. Then, we can dive into the necessary work.


Boeing Commercial Aircraft

Sperry Univac Knowledge Systems Center




  1. As mentioned, my work experience paralleled the advance of computing. This was common for many of my peers in former and later cohorts. My approach always was to find out what was behind the wall or below the floor. Computing is both broad and deep. ML took a focus on the latter. The former is of more importance than people realize. Balance?  
  2. ACM.com is the site for an organization that has been involved with computing from the beginning. It has supported interest groups. One of these deals with computer graphics and has been popular from the beginning.
  3. United Airlines (UAL) Reservations Project – 1966-1970 …
  4. Systems and methods for filtering and smoothing data, US7139674B2, Switlik and Klein

Leave a Reply