BEACON Seminar -The Puzzle of Contextual Biases: Why Can’t We Ignore Irrelevant Information?

Humans are prone to biases when seemingly irrelevant context influences our decisions. For example, imagine looking for a flat in a new city. After finding a very nice house that you unfortunately cannot afford (an irrelevant option), the others might start to look less favourable in comparison. Such biases affect not only complex, high-level decisions, such as which flat to buy, but also basic processes of perception and memory. Researchers can use the highly controlled laboratory conditions available for low-level cognition studies to better understand decision-making in both perception and memory, as well as higher-level cognition. But why do such biases occur? Why can’t we see or remember things accurately even in seemingly simple tasks?
In this talk, I will first discuss theories that aim to understand the mechanisms of contextual biases at different levels of explanation. I will then focus on our recently developed ‘demixing’ model, which predicts that the strength and direction of these biases depend on the amount of noise in the stimuli and their similarity. I will present the results of experiments that tested this prediction in a visual working memory task by manipulating stimuli noise in the colour and orientation domains. The results show that increasing noise not only decreases the precision of responses but also changes the strength and direction of the biases in line with the model’s predictions. According to the model, contextual biases are an inevitable consequence of intermixed noisy signals in the environment, where the most optimal solution is often the biased one. This explanation can also be applied to decisions in higher-level cognition, where one often needs to determine what they prefer while facing many intermixed signals from different information sources. This demonstrates how working memory and low-level cognition can inform our understanding of economic and social decisions, providing an ideal model system for decision-making research.