Autori
Fontana, Magda
Ferraris, Gianluigi

Titolo
Managing knowlwdge in agent-based models: theoretical and methodological issues
Periodico
Università degli studi di Torino. Dip. Di Economia e Statistica Cognetti de Martiis. Working paper series
Anno: 2006 - Fascicolo: 3 - Pagina iniziale: 1 - Pagina finale: 26

The paper proposes an experimental setup to compare different representations of artificial adaptive agents (genetic algorithms, artificial neural networks, and classifier systems) and suggests some criteria to assess equivalence and robustness of performance. In economie theory, the use of artificial adaptive agents as substitutes for the homo oeconomicus raises important methodological issues. While the reductionist approach grounded on Olympic rationality offers full rationality as the unique reference point for problem solving, weaker notions of rationality generate a variety of processes and outcomes of decision-making. The paper gives some suggestions on sensitivity of the behaviour of agents to the algorithmic choice and to the codification of knowledge. Preliminary results show that in an iterated prisoner's dilemma interesting patterns of behaviour (such as strategies that perform better than the tit-for-tat) emerge.



Testo completo: http://www.est.unito.it/do/home.pl/Download?doc=/allegati/wp2006dip/3_wp_fontanaferraris.pdf

Esportazione dati in Refworks (solo per utenti abilitati)