Analysis and synthesis: multi-agent systems in the social sciences

The author thanks Peter McBurney and the conveners of the First International Workshop on Market-Based Control of Complex Computational Systems, Liverpool, September 1–2, 2008, when an early version of this paper was first presented.

This paper draws extensively on Marks (2006, 2007). I wish to thank a referee for his comments.

Another is the influence of physics on early classical economics (Mirowski, 2007).

Indeed, one criticism of simulation might be called the ‘so-what’ or ‘anything-goes’ critique: ‘OK, you've found a model A that results in behaviour B. Now what?’.

Newton knew of Kepler's explanation (from Streete, 1661) when he derived his laws. To what extent was his general derivation triggered by the sufficiency of Kepler's model?

Mirowski (2007) argues that for a 150 years economists have focused on the agents (buyers and sellers) who exchange, and have ignored the structure and procedures of the market in which the exchange occurs. This would explain the lacunæ in the economics literature that confront computer scientists when they seek detailed analysis and explanation of the workings of historical markets in order to implement automated markets of various kinds.

Haefner (2005) lists seven possible goals: usefulness for system control or management, understanding or insights provided, accuracy of predictions, simplicity or elegance, generality (number of systems subsumed by the model), robustness (insensitivity to assumptions), and low cost of building or maintaining the model. Axelrod (2006) also lists seven: prediction, performing tasks, training, entertaining (see those ubiquitous games consoles), educating, existence proofs, and discovery; prediction, existence proofs, and discovery are the main scientific contributions.

Rubinstein (1998) lists four purposes: predicting behaviour, as a normative guide for agents or policymakers, sharpening economists’ intuitions when dealing with complex situations, and establishing ‘linkages’ between economic concepts and everyday thinking. Burton (2003) lists three questions: asking ‘what is’, ‘what could be’, and ‘what should be’.

Midgley et al. (1997) and Marks et al. (2006) describe how they calculate each brand's weekly profits, given the combination of marketing actions of all brands that week, and with prior knowledge of the brands’ costs.

This conceptual framework was introduced by Mankin et al. (1977).

We can see this as referring to elements a_i and model A, respectively, of Section 2 above.

Mirowski (2007) is skeptical about the meaning of this measure when the agents are artificial.

This study follows earlier tournaments involving competing agents, which have resulted in new insights: political scientist Axelrod (1984), marketers Fader and Hauser (1988), and many computer scientists and social scientists in the past 30 years.