How brains learn to control machines

Abstract

After training, animals and humans can make their thoughts interact directly with computers. A study provides evidence that the corticostriatal system of the brain is essential for this learning process. Corticostriatal circuits are known to be involved in physical skill learning, but it is not clear whether these pathways are also important for abstract skill learning. Here, Jose Carmena and colleagues train mice to manipulate an auditory cue towards a goal using activity in the primary motor cortex, but without overt movements. Striatal neurons changed their activity throughout training, with an increasing number modulating their output on the basis of whether the target was reached or not. This modulation and learning was based on standard plasticity mechanisms, as demonstrated when the disruption of NMDA-receptor action eroded the animal's ability to learn the skill. These findings show that corticostriatal plasticity is central to abstract skill learning, and imply that the nervous system's storage and retrieval mechanisms, developed for skilled behaviour, may be transferrable to the control of neuroprosthetic movements.