On Thursday, January 26th at 15:00 hrs we will have another session of our seminar: Denis Bonnay (Université Paris Ouest) will present his work on Trust and Bayes (abstract below).
The meeting will take place in room D1.113 in Science Park 904.
How should an agent take into account the probabilistic opinion of other agents in a group? The traditional Bayesian answer would be to use Bayes’ rule and higher-order probabilities. In 80’s, Lehrer and Wagner proposed a different and much simpler model based on the attribution of epistemic weights, which were meant to express degrees of trust. The interest of the model is to allow for a detailed study of the conditions under which agents may reach consensus by repeated updating on each other’s beliefs. However, the absence of a principled justification for their update mechanisms casted on some doubts on the significance of the results. In this talk, I will discuss whether such a justification can be given and prove a representation theorem for Lehrer and Wagner’s updates with respect to Bayesian updates. (Joint work with M. Cozic)