Content-type: text/html Downes.ca ~ Stephen's Web ~ A recurrent network model of planning explains hippocampal replay and human behavior

Stephen Downes

Knowledge, Learning, Community

Why do we plan? A commonsense explanation based, say, in evolution might be that it is advantageous to plan. As a result, we would have developed that capacity over time. But what does that mean, exactly? What is it to plan, and what are the mechanisms explaining how it offers an explanation. This paper isn't the last word, but it does show how neural network simulations can take us past the hand-waving stage of explanation and offer something more concrete, perhaps pointing the way to a broader pedagogy of planning as well as to the development of AI that can plan. This is a pretty complex technical paper but it will reward the effort. Minimally, it encourages us to think about such things as how long we spend planning, a planning space and planning tasks, a recurrent network architecture of planning, the idea of 'human agent rollouts', and how these resemble what's actually happening in the hippocampus.

Today: 7 Total: 1457 [Direct link] [Share]


Stephen Downes Stephen Downes, Casselman, Canada
stephen@downes.ca

Copyright 2024
Last Updated: May 18, 2024 7:46 p.m.

Canadian Flag Creative Commons License.

Force:yes