LAG

Learning Across Games

 Partecipanti

Mappa

 Word cloud

Esplora la "nuvola delle parole (Word Cloud) per avere un'idea di massima del progetto.

 Obiettivo del progetto (Objective)

'In this proposal we aim to study the twin questions of a) how agents categorize decision problems (and in particular games) according to their similarity and b) how this categorization affects standard predictions in Game Theory. The most important contribution to the literature is that we endogenize the question of categorization, assuming that categorizations can be learned. Existing literature is characterized by making sometimes ad hoc assumptions on categorizations. Initial results have already shown that some of the key concepts used in Game Theory appear to be very fragile once the problem of categorization is introduced. We have also shown that many recent experimental results have a natural explanation in terms of learned categorizations. Possibly because of these results the research has already had quite some impact, reflected for example in a large number of seminar invitations. My personal background allows me to approach this issue in an interdisciplinary manner. Many cognitive scientists have been interested in the question of categorization and there are interesting applications even to Biology, in particular to the problem of genomic imprinting. Other applications in Economics concern bidding of firms in search or procurement auctions. Data from such applications can be used to test the theory. The project is executed at the Department of Economics at Maastricht University, in cooperation with an international group of top researchers working on closely related issues.'

Introduzione (Teaser)

Understanding how the mind works in gaming environments can help researchers explain how we make decisions in complex situations and what we base these decisions on.

Descrizione progetto (Article)

The games we play, from board games to guessing games, are a source of learning for the mind. The EU-funded project 'Learning across games' (LAG) investigated game theory, particularly how the mind categorises decision issues according to their similarity across different games. It also studied how this categorisation affects standard predictions in game theory.

General game theory claims that people endogenise the question of categorisation and assumes that categorisations can be learned. However, the project showed flaws in game theory once the issue of categorisation is introduced. Based on this, it conducted several experiments in gaming to reveal a natural explanation regarding learned categorisations.

LAG published a paper titled 'Learning across games' which showed that distinguishing all games requires too much reasoning resources, causing players or agents to partition the games into categories. In another paper titled 'Learning without counterfactuals', the project examined learning procedures when counterfactuals (payoffs of non-chosen actions) are not observed. It showed how decision makers update their propensities after every payoff experience and transform them into choices, representing a set of natural axioms.

another experiment on learning in a multiple games environment examined how players make decisions if faced with many randomly generated games. It found that while agents extrapolate between games, they learn to play strategically equivalent games in the same way. These trials and conclusions, in addition to a host of others gleaned from the project, have helped increase the understanding of gaming theory, choices and learning in a gaming environment.