Within striatal circuits the neuromodulator dopamine is considered by many to act as a learning signal, controlling neural plasticity (60). One of the main lines of evidence has come from recordings of midbrain dopamine neurons whose axons project to wide regions of forebrain (61)—including the striatum, which has a very high density of dopamine receptors. In macaque monkeys such cells often show brief (~0.1 s) increases in firing rate, particularly to unexpected rewards, or to unexpected cues that signal upcoming rewards (reviewed in 62). The ability of dopamine cells to ignore fully expected rewards and shift firing to cues that predict rewards has a strong resemblance to “error signals” in certain formal learning theories (63) and certain computational models of “reinforcement learning”(64). Both classes of model refer to situations in which the extent of associative learning is controlled by a simple signal that provides feedback on the overall success or failure of an action or expectation. Because the signal does not provide detailed information on the exact nature of any errors committed, it is sometimes described as a “critic” rather than a “teacher” (65). Natural changes in dopamine cell firing rate are typically subtle—an extra spike or two over a tonic rate of a few per second—and often clearly visible only in cumulative records of many trials (66). Such modest changes in dopamine cell firing rate may potentially lead to substantial changes in striatal dopamine release (67). However, increased activity of dopamine cells may also produce no increase in dopamine release in terminal regions. One recent study found that direct electrical stimulation of this cell population would produce increased release of dopamine in ventral striatum, but only if the animal was not expecting the stimulation (68). This implies an exquisite local control over rapid changes in dopamine release that may have as important a part in dopaminergic control of reinforcement learning as altered firing of midbrain dopamine cells. Both mechanisms are likely responsible, in part, for behavioral observations that “the more expected a reinforcer, the less effective it is” (69). Evaluative and anticipatory circuits involving amygdala, hippocampus, and PFC can control dopamine release as part of the normal top-down, cognitive control over learning processes (70,71). (Berke 2003, 81)
References
Berke, Joshua D. 2003. “Learning and Memory Mechanisms Involved in Compulsive Drug Use and Relapse.” In Drugs of Abuse: Neurological Reviews and Protocols, edited by John Q. Wang, 75–101. Totowa, NJ: Humana Press. http://dx.doi.org/10.1385/1-59259-358-5:75.0