"One of the simplest kinds of 'structure' is when the outcome that just occurred tells you something about what is likely to happen next."
"Where people go astray is when they base their decisions on beliefs that are different than what is actually present in the world. In the coin example, if you toss a coin five times and all five times are heads, should you pick heads or tails on the next flip? Assuming the coin is fair, it doesn't matter - the five previous heads don't change the probability of heads on the next flip - it's still 50 percent - but people nevertheless act as though those previous flips influence the next one," added Green.
He said that when things are actually independent over time, meaning they don't have any structure, people will interpret results through possible structures, a way of thinking often seen among gamblers.
For example, gamblers who win three hands in a row, may believe themselves to be "hot" and thus more likely to win the next hand.
Green showed that similar behaviours are seen even in an optimal, fully rational computer learner given similar incorrect beliefs about the world.
In addition, when the context of the task was changed so that subjects understood that the outcomes were actually independent, a drastic shift in their behavior was noted, with subjects all doing the "right" thing for the way the world actually worked.
"This demonstrates that given the right world model, humans are more than capable of easily learning to make optimal decisions," said Green.
The study has been published in the Proceedings of the National Academy of Sciences (PNAS) "Early Edition."
Source-ANI