Post #2384213
2026-04-18 19:55 UTC
@thrilway@kolektiva.social Gotcha! I suppose by "constraints" I primarily mean things like: our visual system perceives a limited slice of EM radiation as an arbitrary color spectrum, or our scent receptors get wired up directly to our memory. However, I think we also have minds constrained to prefer thinking with certain systems of concepts. Like, we perceive scenes with objects and actors, trajectories and intents. I would put "native symbol manipulators" in the same bucket, without necessarily saying we have any innate symbols or even a finite space of grammars.
I certainly agree that AI and CogSci are very different fields with different goals, but I lament that. I wish AI was more concerned with principles of intelligence generally, and how artificial and natural forms of it relate. Especially now, since the most interesting frontiers are how these things intersect. GOFAI was fundamentally flawed, but so are LLMs without logic and facts. Or feelings, for that matter! Another of our innate constraints.
Replies (1)
-
@LyallMorrison@cloudisland.nz 2026-04-18 20:52
@ngaylinn@tech.lgbt @thrilway@kolektiva.social Decades ago for my undergraduate dissertation, I made a demonstration that an agent system using inherited neural network starting weights combined with online learning during the agent lifetime was superior to either approach separately. I captures the idea you mention, that humans and animals are born primed to learn about the kind of world we live in. I believe that's going to be the root of any human-like or embodied AI, because it just works so well.