About Me and the Subjects I Study

I've always been interested in AI. It is, after all, the holy grail of science fiction. I used to daydream about this stuff. I read Asimov, Clark, and Hienlien. I ate up pop-sci books of all sorts and in high school I read "On Intelligence" by Jeff Hawkins on inter-library loan.

When I was a teenager I taught myself to program; I started by reading my father's copy of the C book. First I programmed a maze generator. Then I invented my first AI. It analyzed sentences to predict the next word. It was a Markov model, although I didn't know it at the time. It performed exactly as intended, which is to say the project was completed and then shelved. After that I set my sights on more practical problems: I tried to make equation solvers for my physics homework, but instead I discovered exponential time complexity. I spent a lot of time inventing new and useless algorithms.

I read the source code for the AIs in HalfLife, since it's open source. It turns out that they're all finite state machines. I used to make silly little NN's with garbage training rules. Some evo stuff. This very old work feels a bit cargo culty. I was inspired by the paper "Three States and a Plan: The A.I. of F.E.A.R.," which I found after I played the video game and loved playing with the AI. A few years later I even implemented that algorithm for the bots in NaturalSelection2, although nothing came of that project for lack of dedication.

In college I was excited to take the intro-to-AI course, which they made all compsci majors take. I read the textbook cover-to-cover the summer beforehand. But the textbook (Russel and Norvig) and the course were both stale material. At the time, deep learning was taking off! Artificial neural networks had gone from a toy concept, that no one could get working, to the state of the art and were rapidly progressing. I started reading, and in the college library I could access all of the pay-walled journals. It was exciting to study the academic literature directly, instead of through a teacher or textbook. I printed out everything and kept the papers in a huge binder.

But after a few years I still could not understand how to get deep learning to do what I want it to do. With enough R&D, deep learning can do anything, but that does not make it strong AI. It's still weak AI, and a lot of weak AIs put together does not necessarily make a strong AI. I became disillusioned and I resigned myself to getting a real job.

But then I stumbled on an article that claimed to have a computer model of biological intelligence. Curious, I started watching Matt Taylor clumsily explains basic computer and neuroscience topics in his HTM School video series. I cut classes to watch those videos that day. From then on I've been hooked on computational neuroscience.

The HTM algorithm works and is mathematically rigorous in a way that deep learning systems are not. This makes it accessible to anyone who understands math and computer programs, not only neuroscientists. I had no understanding of neuroscience, but I could unravel this short program to figure out how it worked. Studying this algorithm was my introduction to neuroscience.

Right away I noticed that the state of the art of neuroscience has progressed much farther than perhaps many people think? Growing up the brain was always presented as a great enigma, but that's simply not true anymore. It was here that I realized that it might actually be easier to reverse engineer the brain than to design a new one from scratch, which has been my ambition ever since.