Papers from the AAAI Spring Symposium
To effectively accomplish their goals, agents need to model their environment and other agents with which they interact. Building detailed, accurate, and up-to-date models however is a time-consuming activity and can detract from the actual problem solving activities of the agents. We define "satisficing models" as approximate models that enable agents to reliably perform at an acceptable level of effectiveness.
Agents have to make informed and reasoned decisions about allocating their limited computing, sensing, and other resources toward problem solving versus model building activities. To be able to make these decisions effectively, agents must be able to evaluate the accuracy and reliability of their current models, predict the computational implications of building more accurate models, and analyze which components of their world models will yield the maximum incremental payoff upon enhancement. Research questions to be addressed by the presenters and discussants in this symposium will include the following.
- What are the computational tradeoffs involved in model construction? How can they be measured?
- How to incrementally develop and update the satisficing model with changes in the environment and changes in the collection or behavior of other agents?
- What is the role of inductive learning in resource-bounded reasoning? Should learning be used to control deliberation? How should one control the exploration-exploitation tradeoff?
In addition to presentation of selected papers, the symposia will consist of panel discussions, breakout groups, and invited talks. We will distribute accepted papers and key discussion topics ahead of the symposium.