Agent Concepts and Architectures

Agent Concepts and Architectures

• Agents are computational entities capable of flexible and autonomous activity and interaction in some environment.
• Agents exhibit characteristics such as rationality, mobility, adaptivity, introspection, and benevolence.
• Agents have mental attitudes including belief, knowledge, intention, plan, commitment, desire, and preference.
• Agents and objects are complementary rather than mutually exclusive, with agents allowing for qualitatively different system perspectives and different levels of abstraction.
• Minimally intelligent agents exhibit proactiveness, reactivity, and social ability, and can be difficult to balance goal-directedness with reactivity in complex environments.
• There are different types of agent environments, including accessible vs. inaccessible, deterministic vs. nondeterministic, episodic vs. non-episodic, static vs. dynamic, and discrete vs. continuous.
• Agent architectures include logic-based, reactive, goal-oriented, and layered systems.
• Logic-based agents use symbolic, knowledge-based AI and logical deduction or theorem proving to derive actions based on information about the environment.
• Reactive agents exhibit intelligence through interaction between simple behaviors and direct observations of the environment, with decision-making established through a set of behaviors.
• Goal-based agents evaluate and determine explicit goals that influence action selection, with changing goals depending on the environment response.
• Layered architectures naturally lead to multiple interacting layers of hierarchically organized behaviors, with horizontal and vertical examples.
• Challenges in designing agent architectures include symbol grounding, reasoning time, building a complete model of a complex environment, emergence prediction, and complexity of interactions.