Information Technology Reference
In-Depth Information
humans to use higher-order theory of mind to reason about what other people
understand about mental content, while other animals, including chimpanzees
and other primates, do not appear to have this ability.
A possible explanation for the emergence of higher-order theory of mind is
that higher-order theory of mind is needed in situations that involve mixed-
motive interactions such as negotiations or crisis management [9, 10]. In these
situations, interactions are partially cooperative in the sense that an interaction
can lead to a mutually beneficial outcome, but also partially competitive when
there is no outcome that is optimal for everyone involved. For example, both
the buyer and the seller of a house benefit from a successful sale. However, the
buyer prefers a low sales price, while the seller prefers a high sales price.
In this paper, we consider agent-based computational models to investigate
the advantages of making use of higher-order theory of mind in mixed-motive
settings. We therefore model cognitively more sophisticated agents, in which
there has been increasing interest in recent years [8, 11-13]. These agents perform
actions based on their own desires and goals, but also take into account that the
actions of other agents can influence their situation. By controlling the cognitive
abilities of agents and monitoring their performance, we determine the extent to
which higher-order theory of mind provides agents with an advantage over agents
that are more restricted in their use of theory of mind. We have selected to study
the interaction of cognitive agents in the Colored Trails setting, introduced by
Grosz, Kraus and colleagues [14, 15], which provides a useful test-bed to study
mixed-motive situations. Section 2 describes this setting in more detail.
We compare simulation results of agents of two different types. Agents of the
first type base their beliefs on the iterated best-response. We also consider agents
that use utility-proportional beliefs, which is more consistent with the behaviour
of real life agents [16]. The results from the latter agents should provide insight
in the effectiveness of higher-order theory of mind in mixed teams of agents
and humans, which occur in an increasing number of domains [17-19]. Section 3
describes the two agent types and how these agents make use of theory of mind.
Section 5 presents the results of the simulations. These results are discussed
in Section 6, in which we draw conclusions about whether or not mixed-motive
situations may have contributed to the emergence of higher-order theory of mind
in humans, as well as the extent to which higher-order theory of mind may be
useful for computational agents that interact with people.
2 Colored Trails
To determine the effectiveness of higher-order theory of mind in mixed-motive
settings, we have selected the Colored Trails (CT) setting. Colored Trails is a
board game designed as a research test-bed for investigating decision-making in
groups of people and computer agents [14]. The game is played by two or more
players on a board of colored tiles. Each player starts the game at a given initial
tile with a set of colored chips. The colors of the chips match those on the tiles
of the board. A player can move to a tile adjacent to his current location by
