Intelligent agent strategies for modelling spatial
economies
This paper is concerned with a project in what Epstein
and Axtell (1996) call ‘generative social science’. It involves (op. cit.
p.177) the production of
microspecifications (initial agents,
environments, and rules) that are sufficient to generate the
macrostructures of interest.
The macrostructures in question are patterns of agricultural
land use around a market. Epstein and Axtell endow their landscape with ‘natural’
resources and allow only gathering. What is proposed here is to place a settlement
in an agricultural landscape, to treat that settlement as an arena in which
market interactions can occur, and to have spatially extensive production
activity around it.
There is a substantial gap between the worlds created
by Epstein and Axtell and real geographies. For a start, their worlds have
no production. There is use of labour in the gathering of resources but no
transformation of those resources, except by consumption. As a result, the
landscape is not made by agent activity, apart from the rather limited impact
of gathering. With no production there is little need for division of labour
or for the kind of learning-by-doing that underpins specialisation. The only
economic differentiation between agents relates to their physical capacities
and to the stocks of goods they accumulate. Trade can occur but only through
individual encounters as agents go about the business of gathering. There
are no dwellings, no storage facilities, and so on. In short, there are locations
on the lattice but no places.
Of particular importance for the argument in this paper
is the absence of marketplaces. The reason for this runs deep. Agents operate
in the neighbourhood around their current location, where the neighbourhood
is limited by the agent’s ‘vision’. Agents move within their neighbourhood
to gather resources. They have no capacity for action at a distance (beyond
the neighbourhood boundary). Nor do they have a sense of place other than
of their current location as they have no memory. Furthermore, each cell
in the landscape can be occupied by only one agent. This approach to agent
behaviour in space owes much to the tradition of cellular automata, from
which it inherits some useful features. However, it makes it very difficult
to represent marketplaces and places of other kinds.
So far as we are aware, no one has tackled the connected
problems of production (and land use) decision-making, consumption decision-making,
and market interactions with intelligent agents in a spatial setting. A successful
approach to this problem is a prerequisite for an attack on geo-economic
problems generally, including the problem of producing an urban economic
model (both short-run and long-run). This paper maps out a general strategy
for modelling agent interaction in an agricultural economy around a single
market settlement and presents some early modelling results.
Bill Macmillan
University of Oxford
bill.macmillan@admin.ox.ac.uk
Mike Bithell
University of Cambridge
He-Qing Huang, University of Oxford