A Very Fast Idiot

Photo by cottonbro on Pexels.com

Back in college my major was Polymer Science. Plastics, resins, that sort of thing. Even so, it was deemed necessary for undergrads in technical fields to gain some exposure to computers and programming.

Bear in mind, this was still early days in the computer revolution. Personal computers did not exist, nope, not so much as a Trash 80, though the Apple I was already on its way. What passed for a small computer was a DEC PDP-11, about the size of a large filing cabinet. Paper tape. Punch cards, which were bloody awful. Later, if you were lucky, a dumb CRT terminal. If you weren’t, a paper teletype machine. Our campus system was a XEROX mainframe (Sigma 9), 64K main memory (that’s K, not G or even M). The system took up an entire (and very large) room. Have I dated myself well enough? I should say so.

I knew nothing then, thinking computers were something almost magical a la Star Trek. Not these computers. You had to tell them everything, and I do mean everything, in precise instructions, in order, and they would do what you told them and nothing else. Problem was, what you think you told them wasn’t always what you actually told them, and since we were the last generation running batch jobs it sometimes took a long wait before you knew you’d messed up. They were, as the “elves” in charge of the Sigma referred to them, “very fast idiots.”

I loved it. One of the biggest regrets of my misspent youth was I didn’t change my major in my first year. Regardless, cut to the present. Narrow AI is progressing by leaps and bounds and used everywhere (not always a good thing); general AI is either imminent or impossible, depending on who you ask. I’m taking online classes in machine learning because I can and I want to.

My last lesson was was on gesture recognition. You film yourself on a webcam, and using prefab learning model libraries, teach the computer to recognize a human hand. For practice we created an updated version of Pong, only this time you point at the top of the screen and the computer has to know to put the paddle where you’re pointing to intercept the ball. So far so good, only mine was creating a paddle when my hand wasn’t even in the shot. Took me a moment to realize why: it was interpreting the headstock of my Peavey Predator hanging on the wall as a hand.

Still doing exactly what it’s told, if not exactly what you intended. Still a very fast idiot. I’m not holding my breath on that general AI thing.