Agents often have to make decisions with incomplete knowledge and few computational resources. We argue that qualitative representations and reasoning, especially combined with analogy, provide a natural approach to performing decision-making in situations with little data, incomplete models, and under tight computational constraints. Moreover, qualitative models provide a means of recognizing and framing decision problems. This paper describes our progress in exploring these ideas to date, using examples from experiments with a system that learns to play Freeciv, an open-source strategy game.
|Original language||English (US)|
|Title of host publication||Proceedings of the QR’18|
|Editors||Zoe Falomir, George M. Coghill, Wei Pang|
|Number of pages||7|
|State||Published - 2018|