A Personal Note: AI and I

In 1984, I built my first artificial-intelligence system, although nobody called it that at the time. It cut glass.

Back then, windows were not ordered in finished sizes. An architect submitted a list (so many panes of this dimension, so many of that), and the manufacturer started with large rectangular sheets, eight feet by six feet or larger, and cut the entire order from them. The business lived or died on a single question: What was the smartest way to slice each sheet so that as little glass as possible ended up in the dumpster?

That question is much harder than it sounds. A few dozen window sizes can be arranged on a sheet in an astronomical number of ways, and there is no practical way to examine them all. We did what the AI field was already learning to do: we used heuristics. A heuristic is a rule of thumb, an intelligent shortcut that gives up guaranteed perfection in exchange for a very good answer reached quickly. It allows a system to tackle a problem when checking every possible solution is impossible.

Our program generated a catalog of sensible cutting patterns and handed it to a linear-programming application, mathematical software that selects the best combination from a set of choices. It chose the mix of patterns that filled the architect’s entire order using the fewest full sheets of glass.

The logic worked. The computer did not. On 1984 hardware, our system was slower than an experienced cutter with a pencil and a legal pad. That gap (a sound idea waiting for machinery that did not yet exist) became the story of artificial intelligence for the next three decades. Much of the mathematics behind today’s AI existed long before computers became cheap enough, fast enough, and numerous enough to run it at scale.

By then, I already held two degrees, one in industrial engineering and another in information systems, and I had become convinced that AI would be the next big thing. Right conclusion. Wrong decade.

I left my job and returned to school for a master’s degree in computer science with an emphasis in artificial intelligence. I studied optimization techniques, neural networks (which, unlike a good deal of that syllabus, are very much alive today), pattern recognition, expert systems, knowledge bases, and inference. I learned LISP, the programming language that dominated AI research for a generation.

During summer breaks, I programmed industrial robots at an auto-parts manufacturing plant. Some of those parts were airbags mounted in steering wheels. If I did the same job today, I suspect I would need liability insurance in the hundreds of millions of dollars!

What I really wanted to study was language understanding: teaching a machine to grasp what a sentence means. That ambition is a direct ancestor of today’s large language models, the technology behind ChatGPT and Claude.

I explored the possibility of pursuing a Ph.D. My professors gently talked me out of it. Genuine language understanding, they explained, was still many years away. Realistically, I could make progress on one narrow slice of the problem. At the time, much of the field treated language as probability, predicting which word was likely to follow another. I would have selected one of five or six specialized research topics and spent years on it.

I faced the dilemma that eventually confronts every doctoral student: Do I become one of the world’s leading experts on a very small subject and hope that the subject turns out to matter?

I decided against it and told myself I would wait a couple of years for the field to catch up.

That was 1990. My couple of years have turned roughly thirty.

I had a career in the meantime. I went back to work and worked on expert systems and other information systems applications for Fortune 500 companies including Home Depot, Lowe’s, Macy’s, and Shell Oil. I also installed a run-of-the-mill application for the U.S. Space & Rocket Center’s Space Camp in Huntsville, Alabama. Because the contract was on NASA letterhead, I can say, technically and truthfully, that I once did consulting work for NASA.

I then spent years working on business intelligence and big data applications.

Eventually, I left the computer field and became an entrepreneur. Before long, however, I was building databases for my own businesses and looking for ways to automate labor-intensive work that I was doing myself or paying others to do.

Then, in late 2022, ChatGPT appeared. The thing my professors had said was many years away was sitting in a browser tab, free, waiting for me to type something.

I have used it almost daily ever since: for SQL programming, Excel work, gathering and organizing research, civil-engineering calculations, and, most recently, legal research.

That last one deserves more than a passing mention.

In 2025, I bought a distressed asset — a $250,000 upside, if I could hold onto it. Then a third party sued, and the money was suddenly on the table. Distressed assets are messy. I was staring down the possibility of losing everything without ever getting to say a word in my own case.

So I put ChatGPT and Claude to work. I walked into court representing myself — and walked out with my property. The full story is in Chapter 24.

I spent thirty years waiting for these tools to arrive. You do not have to wait.