What happens in our brain when we are faced with a difficult choice? This is what an algorithm built by machine learning tries to find out. A program that analyzed 10,000 choices until it understood, but also predicted how a human would react.
From psychologists to traders, everyone is wondering how humans will react to a risky choice. Do you prefer to get 50 euros right away, or wait until you have a 50% chance of getting 100 euros? What if we go up to 60%? Or at 1,000 euros? From when do we switch from a safe choice to a more uncertain one with a higher profit? Researchers in neuroscience, psychology and computer science at Princeton University have teamed up to build a program that predicts how a human would react to these dilemmas.
Much research already exists on the subject, but the one published in Science on June 10, 2021 is the largest ever, since no less than 10,000 choices have been integrated into an algorithm which, in the long term, manages to imitate with unprecedented precision. taking a human being’s choice. A real revolution according to the authors: ” We were able to recap historical theories, establish that we can improve them, and find a better representation of the decision-making of the human mind.. “
The machine facing the irrational human
Throughout the 20th century, decision theories have multiplied. The general idea is that the individual, faced with a choice, will select the option that maximizes his benefit, his well-being. In the event that there is an uncertainty, the evaluation is made on a probability calculation known to the user himself. But behind this apparent mathematical solidity hides a whole set of more or less identifiable variables linked to emotions, memory or perception. In short, to the irrationality of humans. So many possibilities that over the decades, no theory has really been able to emerge to reach a consensus.
Yet the demand is great, because beyond the sphere of psychologists, economists and even politicians are very fond of knowing what can make an individual switch from one choice to another. Changes that can lead to very concrete consequences such as the fall of a title on the stock market for example.
Until then, analysis models made directly by humans were much more efficient than those that put machines in control. Up to a point anyway, since the authors claim that their theory works best when there is more data processed. They write : ” This may imply that the complexity of behavioral theories can be constrained by the limitation of data.. In other words, models get finer and more accurate when there are more cases to process, which is better done by a computer than by a human.
Machine learning, an essential component
Here, thousands of humans have therefore “helped” the algorithm to learn decision-making. An apprenticeship that allows him today to imitate at best a human being faced with a risky choice.
Ultimately, the machine describes the entire neural path produced in a human brain when making a decision. For Sudeep Bhatia, a psychology researcher who also wrote an article in Science devoted to the study: “ They have shown the true power of this approach (…) Future work will undoubtedly show advances in many areas thanks to this machine learning method. “
The researcher adds that machine learning has a bright future ahead of it in this type of study: “ This will become an essential component in the toolbox of scientists as it revitalizes, (and perhaps even revolutionizes) theoretical research on human behavior. “
The continuation in video