Information Technology Reference
In-Depth Information
in the world! More specifically in our context of experimentation in creativity, just
as the human designer may use the results of simulating the agent's design in the
game world to modify the teleological model of the agent, at least in principle,
the agent too may use the simulation results to modify its own model and thus
redesign itself. This is
meta-reasoning
[
17
], and in particular
meta-reasoning for
self-adaptation
[
32
].
An agent's potential ability to redesign itself raises both challenges and opportu-
nities for developing a CAD-like environment for designing game-playing software
agents. The challenge is empowering the agent to experiment with its design and
to redesign itself as needed. The opportunity is that if the agent can reflect on its
own design, then, at least in principle, the human designer and the software agent
could cooperate in the redesign process. We call this
interactive meta-reasoning.
We
envision a CAD-like environment in which a human designer begins the process of
interactive meta-reasoning by constructing an initial conceptual teleological model
of the agent's design. The designer may then run a simulation in the game world
to evaluate the design. Once the simulation results are available, the software agent
may reflect on its simulated behavior and on the knowledge and reasoning that led
to the behavior, and refine or revise its knowledge and/or reasoning. The human
designer may intervene at any point in the redesign process, for example, when the
redesign task is beyond the capacity of the agent, until the designer and the agent
collaboratively reach an agent design acceptable to the designer.
In this article, we focus on three core elements of the above vision. Firstly, we
describe the TMKL2 agent modeling language and GAIA's visual editor for enabling
the construction of a TMK model of a game-playing agent in the language such that
the model is both human andmachine understandable. Secondly, we describe GAIA's
architecture for interactive meta-reasoning in which the designer can construct, eval-
uate and revise TMKmodels of the game-playing agent. Thirdly, we describe GAIA's
process of meta-reasoning for self-adaptation in game-playing software agents. We
also briefly relate this work with similar work on agent modeling, meta-reasoning,
game-playing, design thinking, and computational creativity.
17.2 Freeciv, A Multi-player Turn-Based Strategy Game
Our work takes place in the context of a multi-player turn-based strategy game called
Freeciv. Freeciv is an open-source variant of a class of
Civilization
games with sim-
ilar properties. A human player can play Freeciv against one or more opponents,
some or all of whom can be software agents. The aim in these games is to control
and grow a civilization while competing for limited resources against other players'
civilizations. Themajor activities in this endeavor are exploration of the randomly ini-
tialized game world, resource acquisition and development, and warfare, which may
at times be either offensive or defensive in nature. Winning the game is achieved most
directly by destroying the civilizations of all opponents, but can also be accomplished
Search WWH ::
Custom Search