23 October 2012
Post date: Oct 30, 2012 4:27:06 PM
Discussion of different possible agent architectures for representing human behaviour.
- Rule-based systems
- Neural nets
- Classifier systems
- Belief / desire / intention models (from AI, mostly).
- Subsumption architecture (R. Brooks).
- Thresholded rules, pandemonium and related "winner take all" architectures.
- Cost landscapes for geographical movement.
- Scripts
- Genetic programming
- Multi-objective optimization
- Q-learning
- Special purpose hacks
Things to consider when devising or choosing an architecture:
- Perception
- Internal state / memory
- Learning
- Agent maintains a model of the world?
- Models of other agents mental states? (i.e., theory of mind)
- Autonomy
- Inter-agent communication
- Deterministic vs. noisy decision making