In an experiment using a technique called calcium-imaging, which enables the activity of very large numbers of brain cells to be monitored simultaneously and across days, rats were trained to approach one of three different goals on separate days. The correct position changed on a daily basis and in a random manner. The animals were then tasked with discovering which was the correct position that day. As expected, over the route to the rewarded location Place cells were observed as being active – encoding the animal’s instantaneous location.
The new finding was that in a brief decision-making window, before the animal embarked on its journey, the activity of a separate population of neurons could collectively be used to predict the rewarded location that the animal was going, specifically on occasions when the animal chose correctly. Using a machine learning technique called neural manifold learning they found that the neural population activity at the decision-making time and the rewarded location were remarkably similar. Incorrect choices were not predicted suggesting that, when they occurred, mistakes were random.
In particular, the team found that the active neurons mapped out the journey of the correct future path rather than the memory of a previous path. Their results imply that this activity is not simply anticipatory or a recall of immediate past, but rather represent a “mental scenario” of possible choices.