Ask HN: What Are You Working On? (November 2022)
What are you working on? Any new ideas you're thinking about?
Still working on Bosworth, this "box of sensors" project for working with embodied AI. The goal is still to collect fundamental data that represents how an entity "experiences" the physical world around it, with an eye towards trying to make some progress in the area of how it can learn a naive / intuitive metaphysics of the world, naive epistemology, naive physics, and so on. To be even more specific, what I'd love to discover is the answer to a question along the lines of "What's the bare minimum of "stuff" that a learning entity has to come hard-wired with (analogous to whatever human children are hard-wired with through evolution) in order to support the learning of all higher level concepts?"
The other angle I'm hoping to dig into a little bit concerns knowledge representation, and specifically how to represent "things" (whether those things are abstract concepts, concrete entities, or "other") are both the neural / "pattern matching" level AND the symbolic / logical reasoning level. That is to say, I'm interested in the idea of neuro-symbolic systems and how to overcome the "impedance mismatch" that is inherent there.
Am I arrogant enough to think I'll figure all that out? Not necessarily, but these are the areas that I'm exploring because I think these are fascinating topics.
So with all that said, in addition to working on the physical apparatus and associated software for the "box of sensors" thing, I've been going back and doing a lot of reading on metaphysics and epistemology. I also just read Leon Festinger's[2] book A Theory of Cognitive Dissonance, and am increasingly convinced that cognitive dissonance is an important part of how learning works. I can't shake the idea that there's a crucial connection between cognitive dissonance and "belief revision"[3]. I'm also going back to first sources and reading Donald O. Hebb's The Organization of Behavior which contains his ideas on learning that have come to be called Hebbian Learning[4].
Anyway, that's more or less what I'm working on at the moment. The only other thing of note is working on learning/re-learning a lot of maths stuff. That's been an ongoing thing for me for some time, and most recently I went back to a really basic book on Logic (Logic for Dummies actually) to refresh my memory of all of those basic rules of propositional logic and predicate logic. The goal is to move into some set theory stuff soon, and one of those various "How to write proofs" / "a transition to proof based maths" books and work on getting myself up to where I can understand proofs. Not trying to become a mathematician per-se, but just want to understand a lot of this stuff at a higher level than I currently do. And right alongside all of that is an ongoing review of really basic stuff like high-school algebra / trig all the way up through calculus, linear algebra, differential equations, etc. My formal maths education ended with Calc I and that was like 25 years ago, so I have some ground to cover.
[1]: https://news.ycombinator.com/item?id=32967331
[2]: https://en.wikipedia.org/wiki/Leon_Festinger