TCS Daily


Nash Monkey

By Kenneth Silber - April 28, 2003 12:00 AM

You might not expect monkeys to act like Nobel Prize-winning economists. But consider an experiment performed by neuroscientist Paul W. Glimcher and colleagues, and described in his new book Decisions, Uncertainty and the Brain: The Science of Neuroeconomics. The book is an absorbing introduction to the emerging field of neuroeconomics, which combines economic concepts with the study of brains and behavior in humans and animals.

In the experiment, rhesus monkeys were seated facing a computer display. A light at center screen would get their attention, and then other lights would move to left and right. Whether the monkeys looked left or right affected how much Berry Berry Fruit Juice they got as a reward. Looking left was sure to bring a small amount of juice, while looking right could bring a lot of juice or none at all. Meanwhile, activity was monitored in the parietal lobes of the monkey's brains.

The monkeys were conducting an exercise known in game theory as "work or shirk." Looking left was the equivalent of "working" - you show up at the office and get paid, whether your boss checks that you're there or not. Looking right was "shirking" - you skip work and only get paid if your boss doesn't check. The boss, meanwhile, wants to get you to show up while checking as little as possible. The monkeys were playing against a computer that was programmed to be the "boss."

When humans play this game (via computer simulation with cash rewards), they come pretty close to finding optimal rates of working and shirking (or checking and not checking). The resulting situation was identified by Nobelist John Nash (of "A Beautiful Mind" fame) as an equilibrium in which neither player can benefit from changing strategies if the other's strategy stays the same. The optimal rates depend on the payoff matrix (how much cash you get for working, how much it costs your boss to check, etc.).

Interestingly, the monkeys also achieved a Nash Equilibrium. When the payoff matrix was set so that 50 percent "shirking" was the optimal rate, the monkeys looked to the right close to 50 percent of the time. When the matrix was reset so that 90 percent "shirking" was optimal, the monkeys' rightward looks soon fluctuated near the 90 percent level.

Of course, the monkeys had no formal knowledge of game theory. (Neither, for that matter, did most of the humans who had played for cash.) But Glimcher and his colleagues hypothesized that the relevant economic calculations were encoded in physiological activity in the monkey's brains. And indeed, data from monitoring the monkeys' parietal lobes lent support to this idea. The firing rates of parietal neurons fluctuated sharply over successive rounds of "work or shirk," apparently tracking what an economist would call "relative expected utility."

Such empirical results, intriguing as they are, remain fairly thin on the ground. Neuroeconomics is a very young discipline, and how far it will go in explaining human or animal behavior remains to be seen. Still, Decisions, Uncertainty and the Brain makes a strong case that the marriage of neuroscience and economics is a field worth pursuing. Glimcher provides a wide-ranging discussion of neuroscience's history and of the philosophical implications of neuroeconomics.

In particular, Glimcher delves into the longstanding philosophical issue of free will and determinism. He argues that neuroscience has long been hindered by a tendency to view behavior as reflex-like and deterministic, thus ignoring concepts of probability and uncertainty that have played a central role in economics. Moreover, he points out, game theory suggests a particular type of uncertainty that's relevant to the question of free will.

Consider again a game of "work or shirk." The players (human or monkey) are engaged in what at one level seems like a predictable behavior pattern, approximating the Nash Equilibrium. But looked at more closely, there is uncertainty about any individual decisions they make. Indeed, in order to play the game effectively, a player needs to behave in a way that is, as Glimcher puts it, "irreducibly uncertain." If the boss has any ability to predict whether you're coming to work on a particular day, your "work or shirk" strategy is less than optimal.

Similarly, if the computer "boss" has information about whether a monkey is about to look left or right, that monkey is not going to get much Berry Berry Fruit Juice. The upshot is that genuine unpredictability would have carried advantages over the course of evolution. Humans may have developed free will because they needed it to survive.
Categories:
|

TCS Daily Archives