©woodleywonderworks via Flickr
When we implement complex cognitive processes, for example when making decisions, we are subject to cognitive biases. But what about simpler processes like those involved in the most basic learning? In a new study analyzing the data from all existing work on the subject, researchers from Inserm and ENS-PSL show that not only are optimism and confirmation bias present in even the simplest cognitive processes in humans and animals, but so is their integration in learning algorithms would increase their performance. These works, published in Trends in the cognitive sciences suggest that these biases may initially be a very old evolutionary advantage.
Cognitive biases such as optimism and confirmation bias are known to affect our beliefs and decisions. Until recently, it was thought that they were specific to so-called “high-level” cognitive processes, ie those involved in reasoning about complex and uncertain statements. For example, it is known that people overestimate the probabilities of desirable events (France wins the World Cup) and underestimate the probabilities of undesirable events (a marriage ends in divorce).
In a study published in the journal Trends in the cognitive sciences, Stefano Palminteri, Inserm researchers at ENS-PSL and Inserm’s Cognitive and Computational Neurosciences Laboratory, and Maël Lebreton, researchers at the Paris School of Economics, challenge this notion of the impact of bias on optimism and affirmation.
The researchers drew on all the data available in the scientific literature on so-called ‘reinforcement’ learning. It’s a basic cognitive learning process through rewards and punishments that humans share with many animals. From this literature review, it appears that very simple reinforcement learning tests make it possible to highlight behavioral signatures specific to optimism and confirmation biases in individuals exposed to them. These biases appear to be much more widespread than previously thought and are present in even the simplest cognitive processes such as learning to make good decisions through trial and error (reward and punishment).
Furthermore, these prejudices do not seem to apply exclusively to humans: the behavioral signatures also appear in similar animal studies. This suggests that these prejudices arose in evolution from a common ancestor long before the appearance of Homo sapiens, which raises the question of why evolution has selected and maintained what at first glance might appear to be processes capable of producing seemingly irrational behaviors.
Stefano Palminteri and Maël Lebreton believe they have found part of the answer to this question through the results of studies based on computer simulations. These studies have compared the performance of reinforcement learning algorithms – some algorithms incorporate optimism and confirmation bias, while others do not. These simulations show that the presence of confirmation bias in the algorithm actually allows it to learn more efficiently in a variety of situations. These biases could therefore in fact favor survival, which would explain why they have not been corrected during evolution.
The article paves the way for new avenues of research that would help refine our understanding of bias and cognitive processes related to reinforcement learning. In particular, the researchers propose to study the role of these prejudices in the emergence and maintenance of pathological conditions such as addiction or depression. On the other hand, these results suggest that adding these biases to artificial intelligence algorithms could, paradoxically, improve their performance.
#cognitive #distortions #contaminate #simplest #mental #mechanisms