Learning from each other. Formal analysis of multi-agent learning

There is something odd about being a member of a group. On the one hand there are clear benefits: forces can be joined, experiences can be pooled, knowledge can be shared, one can learn from one another (``as iron sharpens iron, so one man sharpens another' as an old proverb says). On the other hand there are downsides: herd behavior, pluralistic ignorance, the bandwagon effect---all cases where personal opinion is overridden by conformity (as in the famous statement of baseball player Yogi Berra about a popular restaurant: ``Nobody goes there anymore, its too crowded'). This project will investigate the complex balance between the advantages and disadvantages of being in a group with a particular focus on learning. A unified approach to single-agent learning, providing a full logical analysis of inductive inference in a setting that combines formal learning theory, belief revision theory, and dynamic epistemic logic, was first developed in the applicants PhD Thesis. Lifting this work to the multi-agent case will demand the development of a formal framework for the modeling of, and reasoning about group learning. One of the main goals of the project is focusing on ways of imposing computability restrictions on knowledge change and belief revision. In this respect the project will contribute to the computational view on learning, by proposing plausible and implementable epistemic models. Overall, the output of the project will be relevant for logic and philosophy of science, but also for developmental psychology, sociology, and economics.