AI and videogames: a “drosophila” for declarative methods

Videogames and research in artificial intelligence (AI) techniques share a fruitful past of reciprocal knowledge exchange. On the one hand, videogames offer unsolved challenges to the academy: these challenges are as hard as those offered by what we can call “serious” applications, yet they can be faced in a controlled and reproducible setting. This characteristic makes videogames a kind of advanced “drosophila” for the researcher in AI: they are the ideal controlled ground in which to invent, experiment, and test new AI paradigms, methodologies and techniques. It is not uncommon to resort to simulated game environments as lesser expensive, yet comparably complex, digital twins. On the other hand, the videogame industry (VI) itself is exemplary of typical demands that AI research cannot meet yet. The VI looks for reduced development costs, better integration with AI tools and real-time performance. Especially, knowledge transfer between developers and AI tools must be as fast and smooth as possible: this is one of the reasons why the machine learning (ML) revolution had so far a controversial reception in the VI, in that ML provides black-box AI modules which are not easily “tunable” and configurable at will, and have non-negligible design-time costs. In order to overcome the above limits, one can consider declarative knowledge representation techniques (DKR). However, some of the shortcomings of DKR methods are fairly challenging to be addressed: performance, ease of use, integration, mining of reusable deductive knowledge are not up to the par yet. All these limitations prevent the real adoption of declarative paradigms in many highly-demanding applicative settings of which the videogame development field is an exemplary generalized testbed. Narrowing the above gaps is nontrivial challenge. In this paper we identify some of the key issues which we deem important for the research community and outline our current research progress.