What is biais? And why are we biased?

Online conference by Gerd Gigerenzer, former director of the Max Planck Institute for Human Development, director of the Harding Centre for Risk Literacy at the University of Potsdam, vice-president of the European Research Council (ERC) and writer-in-residence at the Paris IAS for the month of November 2025, as part of the "Paris IAS Ideas" series.

The ‘Paris IAS Ideas’ online lecture series offers short, stimulating presentations by researchers from the Paris Institute for Advanced Studies, marking the start of their one-month writing residency.

Session exclusively online and in English.
Free upon registration.
Registration is required via the form at the bottom of the page to receive the login link.

Presentation

The research project first analyses the different ways in which the term ‘bias’ is used and the assumptions of rationality underlying each of them. There are apparent contradictions in the evaluation of the same behaviour, for example when relying on information about the base rate is considered rational according to Bayes' rule, but irrational because it seems to reflect biases in the judgement of other people. A second step is to analyse the apparent contradiction between the ‘heuristics and bias’ programme of behavioural economics, in which rationality is assumed to be achieved when bias is zero, and the bias-variance dilemma of machine learning, in which rationality is assumed to require non-zero bias because there is a trade-off between bias and variance (i.e., the sensitivity of a prediction to the particularities of the data sample, a term that corresponds to overfitting). From the perspective of the bias-variance dilemma, a certain degree of bias is a condition of rationality rather than an indication of irrationality.

One hypothesis is that this apparent contradiction can be resolved by introducing the distinction made by Leonard J. Savage and Herbert A. Simon between small (closed) worlds and large (uncertain) worlds. More specifically, the hypothesis is that bias is detrimental only in small worlds where the complete set of future states of the world and their consequences and probabilities are known without any doubt. In all other situations involving a substantial degree of uncertainty, bias may be necessary and functional, even rational.