It's strange and frustrating that most AI researchers don't seem interested in natural intelligence.

Nate Gaylinn

In the early days, when "neural networks" were seen as models of brains, many people seemed at least superficially interested in neuroscience. It's not like that now. I'm sure some folks would say "yeah, and aerospace engineers don't worry about bird flight, either!" but that feels wrong to me.

If all you care about is moving cargo, then sure, flight is solved, and who cares if our designs are "biologically realistic". Similarly, if all you care about is recognizing images, playing video games, and generating slop, then AI is solved. We'll just make the current solutions better.

But I think we've barely scratched the surface of what intelligence actually is! Current AI is so narrow and so shallow by comparison, yet I think people don't even notice that because they haven't actually thought about how intelligent living things are, and in how many different ways!

#science #ai #intelligence

Nate Gaylinn

Not to mention the fact that so many of our current AI solutions are really just models of human intelligence, trained by example. Do we not care about creating genuine, original intelligence? Is bottling it up in a computer so that it's reproducible the same thing as making it in the first place? I'd say obviously not, but apparently many people don't see the difference...

Anthony

@ngaylinn@tech.lgbt If you stuck a human infant into the same "learning environment" as an LLM being trained, they would not learn English. This simple-minded thought experiment demonstrates to me that LLM-based AI, or really anything in machine learning that has this form of training, is not and cannot be a model of human intelligence. Pretending a computer is a human baby doesn't help either but that's another conversation!

I too have been struck by the apparent lack of curiosity of our colleagues. One tentative conclusion I've come to is that there is a subspecies of STEM folks driven largely by impressive demos. I feel like this tendency reflects the perversities of short-termist economic thinking, but in any case the view is a splashy demo that makes it into Nature and the NYT but has shoddy science backing it is superior to excellent science that has no splashy demo. The field "progresses" from one impressive demo to the next. Pollack used to say things along these lines, but that's no surprise given that he founded the DEMO lab: it's right there in the name! I think that mindset is not uncommon, though. Even those who agree the science is important run up against the constraint that there's significantly more funding for flashy demos than for basic research.