Bruno Beaufils

Complete Classes of Strategies for the Classical Iterated Prisoner's Dilemma

The Classical Iterated Prisoners Dilemma (CIPD) is used to study the evolution of cooperation. We show, with a genetic approach, how basic ideas could be used in order to generate automatically a great numbers of strategies. Then we show some results of ecological evolution on those strategies, with the description of the experimentations we have made. Our main purpose is to find an objective method to evaluate strategies for the CIPD. Finally we use the former results to add a new argument confirming that there is, in order to be good, an infinite gradient in the level of complexity in structure of strategies.

Testing Double Auction as a Component Within a Generic Market Model Architecture

Since the first multi-agents based market simulations in the nineties, many different artificial stock market models have been developped. There are mainly used to reproduce and understand real markets statistical properties such as fat tails, volatility clustering and positive auto-correlation of absolute returns. Though they share common goals, these market models are most of the time different one from another: some are based on equations, others on complex microstructures, some are synchronous, others are asynchronous. It is hence hard to understand which characteristic of the market model used is at the origin of observed statistical properties. To investigate this question, we propose a generic model of artificial markets architecture which allows to freely compose modules coming from existing market models. To illustrate this formalism, we implement these components to propose a model of an asynchronous double auction based on an order-book and show that many stylized facts of real stock markets are reproduced with our model.

Towards a dialectical approach for conversational agents in selling situations

The use of virtual agents to intelligently interface with online customers of e-commerce businesses is remarkably increasing. Most of these virtual agents are designed to assist online customers while searching for information related to a specific product or service, while few agents are intended for promoting and selling a product or a service. Within the later type, our aim is to provide proactive agents that recommend a specific item and justify this recommendation to a customer based on his purchases history and his needs. In this paper, we propose a dialectical argumentation approach that would allow virtual agents that have sales goals to trigger persuasions with e-commerces customers. Then, we illustrate the proposed idea through its integration with an example from real-life. Categories and Subject Descriptors I.2.11 [Artificial Intelligence:] Intelligent Agents. General Terms: Algorithms.

L'apport des SMA à la modélisation des marchés financiers

We study in this article several multi-agents models of stock markets. We present in a first part models classicly used by economists to simulate markets, which use a central equation to aggregate agents decisions, which implies that the market is synchronous and non continuous. We show that these kind of models are not sufficient in order to reproduce financial markets behaviors, and present our own orderbook-based market model. We show that reproducing the interaction model used by agents on real markets allows to obtain better results, and allows to better understand how statistical properties observed on real markets emerge from agents interactions through the market.

The Iterated Lift Dilemma

A very small change in the Iterated Prisoner’s Dilemma (IPD) payoff matrix leads to an iterated game called the Iterated Lift Dilemma 1 the properties of which are very different from those of the classical IPD (CIPD). We show that the following ideas are to be noted: (i) two levels of cooperation are now possible, the best one needs a difficult coordination between considered strategiesi (ii) only probabilistic strategies can make a high score when they play against themselvesi (iii) complex dynamics can appear (at the edge of chaos) as soon as three strategies are confronted. Our idea, already argumented in the case of the CIPD, is that, in spite of the model simplicity you can obtain many complex phenomena: it is not true that to be good, a strategy must be simple. Building good strategies for the Iterated Lift Dilemma is then much more difficult than for the CIPD.

Studies on Dynamics in the Classical Iterated Prisoner's Dilemma with Few Strategies

In this paper we study Classical Iterated Prisoners Dilemma (CIPD) dynamics of pure strategies in a discrete and determinist simulation context. We show that, in some very rare cases, they are not quiet and ordered. We propose a classification of ecological evolutions into categories which represent complex dynamics, such as oscillatory movements. We also show that those simulations are very sensitive to initial conditions. These experimentations could call into question classical conclusions about interest of cooperation between entities playing CIPD. They may be used to explain why it is not true that cooperation is always the convergent phenomenon observed in life.

Tricher n'est pas jouer ! Simulation et Théorie des jeux: problèmes méthodologiques

Computational game theory is a way to study and to evaluate agents using models from game theory and computer simulations. One of the well known examples is the classical iterated prisoners dilemma (CIPD). It allows the construction not only of complex behavior but also the evaluation of their strength. Setting up such evaluation using classical games situation is not as easy as it seems. We show methodological issues which have to be taken care of, or avoided in order to prevent trouble in simulation results interpretation. Based on some simple illustration, we exhibit two kinds of bias that could be introduced with some idea of how to fixed them.