One of the things I've been wondering about lately is whether advances in computational power, AI, and interface have diminished this made this problem, or made it greater.
On the one hand, we can make both the game interface and the opponent AI much more sophisticated than ever before. Driven by the multi-billion dollar commercial gaming industry, this continues to develop by leaps and bounds.
On the other hand, when simulations look like simulations (as with any board game), users can also more easily recognize--and potentially consider and debate--the assumptions that are built into the game design. That's less likely to occur, I think, as the sophistication of a computer game increases.
Whether this matter depends to some extent on what we're modelling. If it is straight force-on-force, the physics and Pks and so forth have been well understood by the OR folks for years. When we get into social dynamics—so essential to most COIN/stabilization scenarios—its all rather more indeterminate. In those cases, I think there's a real danger of increasingly sophisticated simulations passing off as "fact" what is essentially not very well understood.
This is an argument that one sometimes hears in the physical and design sciences--that for all its remarkable contributions, for example, CAD has also come at a cost in the quality of architectural production. (For those who are interested in the critique, see Sherry Turkle's Simulation and its Discontents). As we develop increasingly sophisticated COIN simulations, and try to capture the complex political behaviour of actors with derived rules or algorithms, there any risk of the same sort of problems?
I don't have a firm position on the issue, but i do think its an interesting set of questions...