From neural activity to movement
The human brain has about 85 billion neurons, of which about 15 billion are part of the cortex. These cortical neurons ﬁre action potentials (spikes) at a rate of about 10 Hz. Thus, neurons in our cortex emit about 150 billion spikes per second. This activity underlies our sensory perceptions, our thoughts, our decisions, our actions. One of the central problems of systems neuroscience is that of decoding these vast spatial and temporal patterns of neural activity so as to interpret them and assign meaning to them. In the last ten years, much progress has been made in decoding activity of neurons in the motor cortex, an output area of the brain that controls movement through its projection to muscles via the spinal cord. In this talk I will report on reproducible experiments carried out by several groups since 2000. These experiments are based on the implantation of multielectrode arrays that record neural activity in awake behaving monkeys. The arrays allow us to monitor the activity of about one hundred neurons in motor cortex during the execution of sequences of reaches to provided targets. I will describe our theoretical eﬀorts to construct models that capture the underlying relationship between neural activity and movement, and thus predict the direction and extent of a reach before the movement is executed. From a theoretical point of view, these studies have allowed us to make substantial progress in our understanding of the neural code. From a practical point of view, our increasing ability to extract information from neural signals has allowed us to translate neural activity into commands to control computer cursors and robotic manipulators. The potential of this approach to restore motor behavior in severely handicapped patients motivates pioneering interdisciplinary research in Brain Machine Interfaces (BMIs), a new area at the frontier of systems neuroscience.