Agent-Based Models for Higher-Order Theory of Mind - Advances in Social Simulation - page 210

Agent-Based Models
for Higher-Order Theory of Mind

Harmen de Weerd, Rineke Verbrugge, and Bart Verheij

Institute of Artificial Intelligence, University of Groningen, Groningen, Netherlands
[e-mail] , { rineke,b.verheij } @ai.rug.nl

Abstract. Agent-based models are a powerful tool for explaining the
emergence of social phenomena in a society. In such models, individual
agents typically have little cognitive ability. In this paper, we model
agents with the cognitive ability to make use of theory of mind. People
use this ability to reason explicitly about the beliefs, desires, and goals
of others. They also take this ability further, and expect other people to
have access to theory of mind as well. To explain the emergence of this
higher-order theory of mind, we place agents capable of theory of mind in
a particular negotiation game known as Colored Trails, and determine
to what extent theory of mind is beneficial to computational agents.
Our results show that the use of first-order theory of mind helps agents
to offer better trades. We also find that second-order theory of mind
allows agents to perform better than first-order colleagues, by taking
into account competing offers that other agents may make. Our results
suggest that agents experience diminishing returns on orders of theory
of mind higher than level two, similar to what is seen in people. These
findings corroborate those in more abstract settings.

1 Introduction

In everyday life, we regularly interpret and predict the behaviour of other people
by reasoning about what they know or believe. This theory of mind [1] allows
us to understand why people behave a certain way, to predict future behaviour,
and to distinguish between intentional or accidental behaviour. People also take
this ability one step further, and consider that others have a theory of mind as
well. This second-order theory of mind allows us to understand sentences such as
“Alice doesn't know that Bob knows that she is throwing him a surprise party”,
by attributing to Alice the ability to have beliefs about Bob's knowledge. In this
paper, we make use of agent-based computational models to explain why our
ability to reason about mental content of others may have evolved.
The human ability to make use of higher-order (i.e. at least second-order)
theory of mind is well-established, both through tasks that require explicit rea-
soning about second-order belief attributions [2, 3], as well as in strategic games
[4, 5]. However, the use of any kind of theory of mind by non-human species is a
controversial matter [6-8]. These differences in the ability to make use of theory
of mind raise the issue of the reason for the evolution of a system that allows