Interactive Meta-Reasoning: Towards a CAD-Like Environment for Designing Game-Playing Agents - Computational Creativity Research: Towards Creative Machines - page 356

Molineax and Aha [ 50 ] describe TIELT, an environment for evaluating learning in
game-playing agents.
Proposals for developing CAD-like environments for games date at least as far
back as Nelson and Mateas [ 58 , 59 ] and Jones et al. [ 38 ]. Nelson [ 57 ] describes
an interactive game-design environment that supports the game designer by making
knowledge of the game mechanics explicit; in particular, it uses first-order logic to
capture the mechanics in video games. In contrast, our work focuses on designing
game-playing agents; thus, it supports the designer by making the knowledge and
the reasoning of the game-playing agents explicit. Both lines of work however share
the goal of game/agent design at the knowledge-level [ 60 ] as opposed to solely the
behavioral level.
For complex game worlds, such as that of multi-player turn-based strategy games,
it is not possible in general to design a game-playing agent such that it always selects
the right action in any state and thus always achieves its goal or maximizes its utility.
This is because multi-player, turn-based strategy games are dynamic, only partially
observable, non-deterministic, and have huge state spaces. The dynamic behavior of
such a game emerges out of the interactions of its numerous, heterogeneous compo-
nents. Instead, game-playing agents can, and often do, fail to achieve their goals in
such game environments. In such cases, retrospective agent adaptation is needed.
Further, if a game designer modifies the mechanics of a game without also revising
the software agents against whom the human player plays the game, then the game
may lose balance and the human player may lose interest. In such cases, proactive
agent adaptation is desirable.

17.5.4 Design Thinking

The term design thinking has become quite popular over the last generation in part
because of its adoption at IDEO 5 and Stanford University's Design School. Design
thinking generally refers to the cognitive processes engaged in design [ 18 , 19 , 21 -
23 , 28 , 30 , 31 , 62 , 64 , 68 - 71 ]. The cognitive processes in design thinking include
not only the memory, reasoning and learning processes in a designer's mind, but
also the cognitive processes situated in and distributed over physical, informational,
organizational, and social worlds.
For the purposes of this chapter, we characterize design thinking as follows:
(1) Design thinking pertains to problems that initially are ill defined [ 64 ].
(2) Design thinking is iterative, entailing experimentation with multiple designs,
with each iteration resulting in a deeper understanding of the problem as well as
the space of solutions, and engaging both problem reformulation and redesign
[ 21 ].
(3) Design thinking is reflective, entailing the construction of cognitive artifacts such
as drawings and journals that facilitate sharing, conversation, and, criticism [ 69 ].

5 www.ideo.com