Key words :
The Neuroscience of Decision-making
5 Jul, 2007 11:54 am
We often make decisions based on imperfect information from a variety of potentially conflicting sources. A new experiment reveals how monkeys and their brains solve such problems. By measuring the electrical activity from neurons in the brain, we discovered how the brain combines information to make decisions: it calculates and combines probabilities in a manner similar to a statistician.
This dilemma is similar to many decisions — big and small —that we make in our lives. They are based on kind of prediction about the state of the world, or another person’s action, or even our own happiness, and they necessitate a form of probabilistic reasoning. Often there is no right or wrong answer, but a better or worse choice based on the available evidence. This kind of questions seems to be rather difficult, but our brains routinely solve them. We can make good decisions based on inaccurate information, sometimes without being even aware how we achieve it. How does the brain do it? Answering this question will provide key insights into the mechanisms that underlie a variety of mental capacities that neurologists call “higher cortical function” or, more loosely, cognition. As the available technologies still lack the ability to provide enough details about our own brains, we must turn to animal research for answers.
Now, the question arises: Is it possible to study probabilistic reasoning in an animal that lacks the full human repertoire of cognitive functions? The answer in this case turns out to be, yes. We trained rhesus monkeys to perform a task very similar to predicting the weather. In the monkey version of weather prediction, four randomly selected simple shapes were shown on a computer screen. Afterwards, the monkey had to make a decision between a pair of colored targets—one red, the other green. The monkey simply moved its eyes to indicate its choice. Only one of the targets was associated with a reward. Instead of predicting the weather, the monkey predicted whether the red or the green target contained the reward.
The shapes play the role of the evidence sources. Some favor red; some favor green; some are more reliable than others. In total, there were 10 different shapes, each conferring a different degree of probabilistic evidence. On each trial, the monkey saw four shapes — that is, one out of a total of 10,000 possible shape sequences. Sometimes the shapes provided overwhelming evidence for one of the targets. But on many trials, the shapes contradicted each other and favored one choice over the other only weakly. Also, just like the weather reports, the evidence does not guarantee the outcome that it favors. For example, the evidence could favor green, but there is still a small chance that the reward might nonetheless be present at red.
To make a good guess, the monkeys needed to learn how accurate these shapes were and how to combine their probabilities. The only way for them to learn was through trial-and-error.
Amazingly, after several months of training, both monkeys were able to learn the task. They were like probability calculating machines. We analyzed their behavior and found out that they were able to learn the probabilities of each shape and combine them to make a good decision.
Furthermore, when the monkeys were performing the task, we monitored the electrical activity of neurons in their brain. Neurons are cells in the brain that carry out most of the calculations. They do this by making electrical pulses, termed action potentials, which are transmitted to other cells through the release of chemicals (neurotranmitters) at their contacts to other cells (called the synapse). We can measure the pulses from single neurons in the brain by placing a tiny wire—a microelectrode— next to these cells. The microelectrode is similar to the brain electrodes used in human neurosurgery for Parkinson’s disease and epilepsy monitoring. Only, our electrodes are much finer. Because the brain itself has no pain receptors inside it, these electrodes are not felt by the patient or the monkey.
We put our microelectrodes in brain region called lateral intraparietal cortex (area LIP). In humans, this brain region is a part of the association cortex that plays a role in awareness, understanding, calculating, attending, planning, and decision-making. In the monkey, this brain area is known to play a role in decision-making, particularly when the decision involves a choice about where to look.
While the monkey viewed the evidence from the shapes, we discovered that the neurons in area LIP changed the rate of their pulses. This “firing rate” encoded the probability associated with the shapes that monkey were viewing. In fact the neurons kept a running tally of the sums and differences of these probabilities (in fact they added logarithms of the probabilities). When a new shape was shown, the neuron’s activity was updated to reflect the new reward probability. This observation is consistent with the idea that the brain calculates probabilities during the decision process.
By recording from the brain, we also learned something about why choices are not always consistent. What is responsible when a poor choice is made? It is known that a neuron’s response (the rate of pulse production) can vary slightly even under the exact same condition. When evidence is strong, such small variation hardly matters. But when the evidence is weak, a slight change in the neuron’s activity may swing the final decision toward the wrong direction. This is exactly what we see in our experiment. The monkeys performed almost perfectly when the evidence was strong, choosing the target that was more likely to give reward. But when the evidence was weak, they sometimes failed to make the better choice. By comparing their choices with the measured neural activity, we found that some of their mistakes could be attributed to the variation in the brain activity.
These experiments show us how the brain calculates probabilities and combines them to achieve a simple form of probabilistic reasoning. Although the experiments were done with the monkeys, they take us one step further to understand the cognitive processes in our own brains.
Yang, T. & Shadlen, M. N. Probabilistic reasoning by neurons. Nature 447, 1075–1080 (2007).
Key words :
2nd paragraph: "They are based on kind of prediction" should be "...based on a kind..."
This is a well-explained review of a very important piece of work. The discovery that neurons in our brains might be calculating probabilities is brings us one step closer to understanding free will, or the lack thereof.