Sensorimotor Intelligence: The Thousand Brains Pathway to More Human-Like AI

Miscellaneous
1

A nice blog post from Greg Robison

4 Likes
2

Thanks! I’m the author of the blog post and just joined the discourse group to learn more. Looks like an amazing group.

4 Likes
3

Welcome @grob and thanks for the great article!

Here’s a post about community engagement - Community Involvement 🌏 to help get oriented.

1 Like
4

Hi @grob!
I have read your article on Medium Active Learning Machines: What Thousand Brains Theory and Piaget Reveal About True Intelligence (@vclay referenced it in the recent video). There you said/asked:

I’m not sure how to add “motivations”, but we’ll also need some kind of computational equivalent to an infant’s innate curiosity, which drives active exploration and experimentation without explicit external rewards, guiding the system to seek out novel, informative experiences that maximize learning

I left a comment there and I’m curious whether others here share this understanding or see it differently):

Actually, a function of “quasi-curiosity” is already implemented in Monty. By its very design, the system tries to extract objects from its environment and understand their interrelationships. This is realized through the Goal State Generator mechanism within the Learning Module (LM), where the goal state involves reducing the agent’s uncertainty about its environment.

Thus, the system genuinely acts like a child exploring the world. It strives to understand, figuratively speaking, “what’s around the corner?”, and this serves as its permanent motive. What truly amazes me about this is that this system’s behavior is emergent. That is, it’s a consequence of executing very specific technical tasks, but, in my opinion, it leads to extraordinarily significant implications. In my view, Monty effectively follows the Algorithm of Reason, whose purpose is to comprehend the universe. It never stops, and the same thing happens with Monty!

1 Like
5

Thanks for this insight! I’ve been reading the latest TBP Monty articles and understand the architecture much better now and agree that a key module provides goals, that can be analogous to curiosity. I agree that it is a key piece of motivation for unsupervised learning.

2 Likes