To start off, there are a couple of different directions the research goes in. Studies tend to be inspired by dissociations in memory performance between Parkinson's patients and medial-temporal lobe (MT) amnesics. The dissociation is often framed in terms of 'declarative' vs. 'non-declarative' memory systems.
In these dissociations, the question of reward is often important. Speaking very loosely, Parkinson's patients don't seem to be able to learn the connections between their actions and reward, while MT patients seem to be unable to learn abstract facts about the world that can be related verbally. (Knowlton et al. 1994)
Learning through reward is often seen as a human implementation of what in Computer Science is called Temporal Difference (TD) learning. This involves making predictions about outcomes, taking actions on the basis of those predictions, and then checking the difference between the prediction and actual result (the error term). Very impressive recording work on single dopaminergic cells in monkeys suggest that dopaminergic cell firing encodes this error signal. (Montague, Hyman, Cohen 2004)
Researchers believe that these findings have serious implications for learning, categorization, and memory. One major implication of the TD/dopamine finding is that learning may be much more tied to action than is often implied by traditional views of learning and categorization. But there are many different ways to look at the basic findings, and I'll be dedicating future blog posts to the different sorts of perspectives people put forward.
References:
- Montague, Hyman, & Cohen (2004). "Computational roles for dopamine in behavioural control." Nature. 431 (7010): 760-767
- BJ Knowlton, LR Squire, & MA Gluck (1994). "Probabilistic Classification Learning in Amnesia." Learning and Memory. 1: 106-120.
