Themes
Dealing with NP hard problems and uncertainty.
Tasks which may befall the intelligent agent
Framing goals. Planning actions to achieve goals; searching. Storing knowledge. Inference. Learning.
Building mechanical robots. Making actuators. Manipulation. Vision and sensing; feature extraction.
Interacting with humans. Safety.
Characterization of research effort
Experiment on real data: observe flaws of algorithms; improve performance by tweaking them.
The intelligent agent
Interaction model
An abstract agent can be thought of having actuators (to perform actions) and sensors (to gain percepts).
Feedback to the agent may be immediate or deferred.
Environment traits
The environment an agent operates in may be characterized based on its observability, stochasticity, episodes, static/ dynamic, discrete / continuous, adverseries.
Rationality
Rationality, bounded rationality, satisficing. Performance measure. \tbc
Attitude towards risk
See decision theory in statstics ref, game theory ref to observe risk evaluation. But, an agent may not always go for the option with the least expected risk, but for the option which yields some low risk whp. This is risk averseness, observed in humans.
Models of rational agents
Simple reflex agents with condition/ action rules. Such an agent is essentially procedural.
Model based agents maintain an internal state (which depends on the actual state of the environment). The agent then does declarative reasoning to determine goals and actions.
Performance measures
Goals, utility function. Learning agents.
Exploration vs exploitation tradeoff. \tbc