"this brain has a mind of its own"
May. 5th, 2009 01:51 pmMarvin Minsky talks about cognition as "society of mind"... which replaces the homunculus with a society of homunculi. The point is to think of the brain as a collection of agents working in parallel, each with a "mind" of its own. Cognition arises from the parts plus their communications with each other.
Some people identify agents with modules, and there are many philosophical-ish debates about what this means (Fodor's "Modularity of Mind", Pinker, Spivey's "Continuous Mind"). One can also wonder how agents/modules relate to functional and anatomical connectivity between brain regions.
Anyway, it looks like multiple levels of the brain do reinforcement learning (by which I mean decision-theoretic reasoning and planning towards the goal of achieving delayed rewards). Although the planning capacity may be shared by different agents, I would suspect that each agent has its own reward function, and thus different goals... which is why "conflict resolution" is a necessary component of cognitive systems.
For example, imagine an addicted person who wants to quit: some agent disagrees with the higher level consciousness. This may boil down to different discount rates: the craver wants to "feel" good right now, whereas the conscious person wants to feel good over the next year. But regardless, it's not clear how one can control the lower levels of the mind.
Achieving self-control: can one align the goals of the higher levels with those of the lower level agents? Are there schools of self-help based on these ideas?
Tangential: Many agents, including many instinctive ones (fear of snakes, sexual attraction), depend on learned high-level percepts (newborns learn to recognize snakes over some time). The fact that we can't turn off these instincts means that we don't have direct control over the higher levels of the visual cortex.
Some people identify agents with modules, and there are many philosophical-ish debates about what this means (Fodor's "Modularity of Mind", Pinker, Spivey's "Continuous Mind"). One can also wonder how agents/modules relate to functional and anatomical connectivity between brain regions.
Anyway, it looks like multiple levels of the brain do reinforcement learning (by which I mean decision-theoretic reasoning and planning towards the goal of achieving delayed rewards). Although the planning capacity may be shared by different agents, I would suspect that each agent has its own reward function, and thus different goals... which is why "conflict resolution" is a necessary component of cognitive systems.
For example, imagine an addicted person who wants to quit: some agent disagrees with the higher level consciousness. This may boil down to different discount rates: the craver wants to "feel" good right now, whereas the conscious person wants to feel good over the next year. But regardless, it's not clear how one can control the lower levels of the mind.
Achieving self-control: can one align the goals of the higher levels with those of the lower level agents? Are there schools of self-help based on these ideas?
Tangential: Many agents, including many instinctive ones (fear of snakes, sexual attraction), depend on learned high-level percepts (newborns learn to recognize snakes over some time). The fact that we can't turn off these instincts means that we don't have direct control over the higher levels of the visual cortex.