“The brain reaches a decision by combining samples of evidence in much the way a good statistician would,” says Michael Shadlen, a Professor of Neuroscience at the Kavli Institute for Brain Science at Columbia University. In a new paper in Neuron, Shadlen and colleagues from the University of Washington and the Shanghai Institutes for Biological Sciences demonstrate this theory by monitoring the decision-making process in rhesus monkeys to determine how much and what information they need to confidently choose a correct answer.
The monkeys were shown a sequence of shapes that served as clues about the location of a reward. They could look at as few or as many such clues before making their choice. The scientists found that the monkeys’ neurons increased or decreased their activity depending on whether the shape in a sequence supported one or the other location (or color). The process halted when the accumulated evidence reach a critical level. This strategy explained both the choice and number of shapes used to make it.
“It’s the brain doing a statistically optimal procedure,” Shadlen says. “It’s nothing less than a basis of rationality. The brain allows us to combine apples and oranges and lemons, so to speak, by assigning them the right kinds of weights so that when we put them together we reason according to the laws of probability.”
This statistical way of decision making resembles a process Alan Turning’s team did in Bletchley Park, England, to work out the settings of German enigma machines. In order to make use of the large clicking machine — called ‘Christopher’ in the recent historical drama “The Imitation Game” — Turing’s team analyzed pairs of randomly intercepted German messages, aligned them one above the other to accumulate evidence from letter pairs (matched or not) until they reach a threshold level of certainty that the messages were sent on identical enigma machines, or not. Once the threshold was reached, the code breaker would either accept or reject the hypothesis.
“Rejections were common, but acceptance allowed them to take the next step, using the brute force method of the machine toward figuring out the message used to establish the settings of the enigma machine,” Shadlen says. “And if they had that, then all codes could be broken that day.”
Kira et al., A Neural Implementation of Wald’s Sequential Probability Ratio Test, Neuron (2015), http://dx.doi.org/10.1016/j.neuron.2015.01.007