When I'm typing, sometimes I hit in between two keys, or I hit the wrong key entirely, and before I know it, I'm already reaching for the delete key. I'm sure you do it too.
Hidden in this ordinary action is a rather impressive trick of the nervous system. When I hit between two keys, sensory cells in my skin and my finger muscles detect that something's gone awry. Then the signal travels up sensory nerves, into my spinal cord, and ultimately up into the motor cortex of my brain. The signal travels quickly, but it still takes 50 or 100 ms before my brain can receive the signal, process it and send out a corrective reaction.
As I type pretty quickly – I timed it at around 5–10 characters per second – the 100 ms it takes to process my typing error is actually an unacceptably long time. So, rather than typing (or correcting) slowly, my nervous system has learned to predict where my fingers are going to be. Based on the signals going out to my finger muscles and the sensory inputs coming in (which actually reflect the sensations from 50 ms or so earlier), cells in my brain can predict that I'm going to miss a key and start correcting early. By the time the cutaneous sensation of a missed key actually makes it up into my brain, it just serves to confirm what these cells had already predicted.
We've known that cells in the cerebellum seem to predict the future this way, but new research from Michael Dimitriou and Benoni Edin shows that the predictions may take place at a much lower level.
They examined the output of muscle spindles, which sense the contraction speed and acceleration of muscles, in healthy human subjects performing a typing task. The subjects typed a sequence like 5, 3, 5 on a numeric keypad, always beginning and ending with 5, which is in the dead center of the keypad. A subset of the muscle spindles, called type Ia afferents, fired at a rate that was proportional to the speed of their muscles, but at a time around 160 ms in the future.
Muscle spindles sense both speed and acceleration, and Dimitriou knew that speed and acceleration together can be used to make a fairly good prediction of future speed. So, he tested whether the predictions from the type Ia afferents were better than just extrapolating from current speed and acceleration. In fact, the Ia afferents performed substantially better, predicting about 80% of the variance in speed 160 ms in the future, whereas extrapolating speed and acceleration matched less than 50%.
Muscle spindles don't just send information up to the brain; they also receive descending inputs. The Ia afferents seemed to be incorporating some extra information coming from the brain – perhaps about the states of other muscles in the fingers – to make a better prediction. It seems, based on the researchers' findings, that those inputs must contain some information about what neighboring muscles are doing.
Interestingly, this means that when we're first learning to type – or perform any new motor behavior – our brains need to learn the task, but maybe our muscle spindles need to learn it too.