340 likes | 1.07k Views
Agents. Today: What Is an Intelligent Agent? Types of agents: Simple reflex agents Agents that keep track of the world Goal-based agents Utility-based agents Environments. What is an agent? . AGENT - perceives with sensors acts through effectors. . The rational agent .
E N D
Agents • Today: • What Is an Intelligent Agent? • Types of agents: • Simple reflex agents • Agents that keep track of the world • Goal-based agents • Utility-based agents • Environments 59.302 Artificial Intelligence
What is an agent? • AGENT - perceives with sensors • acts through effectors. 59.302 Artificial Intelligence
The rational agent A rationalagent :- performs correct actions Performance measure -- the criteria for determining success of an agent is. • Example. vacuum cleaning agent. (see text book and notes) • Plausible performance measure: • the amount of dirt cleaned in a certain period of time. • Rationality is not omniscience: 59.302 Artificial Intelligence
The rational agent • What is rational at any given time depends on four things: • The performance measure • Perceptions so far (the percept sequence) • Knowledge about the environment • The actions the agent can perform 59.302 Artificial Intelligence
The ideal rational agent • Ideal rational agent • action maximises its performance measure, • using evidence percept sequence • using built-in knowledge. 59.302 Artificial Intelligence
Ideal mapping from percept sequences to actions Table of the perception-action (For most agents, this would be an infinite table.) A mapping from the percept sequences to actions. Specifying which actions for any percept sequence provides a design for an ideal agent. 59.302 Artificial Intelligence
Autonomy Autonomous system if determined by own experience. Preprogrammed with no attention to its percepts:- agent lacks autonomy. An agent's behavior can depend both on its built-in knowledge and its experience. The most successful agents have some built-in knowledge and also the ability to learn. 59.302 Artificial Intelligence
Structure of intelligent agents • The agent program: mapping from percepts to actions. • Assumption: agent=architecture + program • The architecture • makes the percepts from the sensors available • it runs the program and • it feeds the program's action choices to the effectors 59.302 Artificial Intelligence
Structure of intelligent agents • Designing points: • possible percepts and actions • goals or performance measure • environment it will operate in. 59.302 Artificial Intelligence
Agent programs • All agent programs: • accept percepts from the environment and • generate actions. 59.302 Artificial Intelligence
The table lookup 59.302 Artificial Intelligence
The table lookup Why not do table lookup? Chess agent needs a table with about 35100 entries. No autonomy at all :-calculation of best actions is entirely built-in. Environment changes:- agent would be entirely lost. Learning with large table is difficult. 59.302 Artificial Intelligence
The taxi driver agent • The full taxi driver task is extremely open-ended • there is no limit to the novel situations that can arise. • What are: • the percepts? • the actions? • the goals? • Performance measures? • Operating environment? 59.302 Artificial Intelligence
The taxi driver agent Performance measures Getting to the correct destination, minimizing fuel consumption and wear and tear, minimizing trip time and cost, minimizing traffic violations and disturbances of other drivers, maximizing safety and passenger comfort. Operating environment City streets? highways? snow and other road hazards? driving on right or left? 59.302 Artificial Intelligence
Types of agent programs • Simple reflex agents • Agents that keep track of the world • Goal-based agents • Utility agents 59.302 Artificial Intelligence
Simple reflex agents Example: factorial agent Condition-action rule: If condition then action 59.302 Artificial Intelligence
Simple reflex agents Condition-action rule: If condition then action 59.302 Artificial Intelligence
Agents that keep track of the world • Updating the state requires the agent to have information about: • how the world changes over time • how its actions effect the world. 59.302 Artificial Intelligence
Agents that keep track of the world 59.302 Artificial Intelligence
Goal-based agents 59.302 Artificial Intelligence
Utility-based agents 59.302 Artificial Intelligence
Environments • Accessible vs. inaccessible: • sensors detect all aspects of environment relative to choosing an action • Deterministic vs. nondeterministic: • is the next state entirely determined by the previous state and action • Episodic vs. nonepisodic: • - episode = (percept,action) pair • - episodic = episodes do not depend on actions in previous episodes • Static vs. dynamic • Discrete vs. continuous 59.302 Artificial Intelligence
Summary The major points to recall are: An agent perceives through sensors and acts in an environment. An agent is made up from an architecture and an agent program. An ideal agent maximises its performance measure, given the percept sequence it has seen so far. An agent is autonomous to the extent that its action choices depend on its own experience. Reflex agents respond immediately to percepts, goal-based agents act so that they will achieve their goal(s), and utility-based agents try to maximize their own “happiness”. 59.302 Artificial Intelligence