Four basic types in order of increasing generality:
Simple reflex agents
Model-based reflex agents
Goal-based agents
Utility-based agents
Showing posts with label agent. Show all posts
Showing posts with label agent. Show all posts
Sunday, 18 January 2009
Rational agents
An agent should strive to "do the right thing", based on what it can perceive and the actions it can perform. The right action is the one that will cause the agent to be most successful
Performance measure: An objective criterion for success of an agent's behavior
E.g., performance measure of a vacuum-cleaner agent could be amount of dirt cleaned up with penalty for amount of time taken, amount of electricity consumed, amount of noise generated, etc.
Better: One point for a clean square at each time unit
Rational Agent: For each possible percept sequence, a rational agent should select an action that is expected to maximize its performance measure, given the evidence provided by the percept sequence and whatever built-in knowledge the agent has.
Rationality is distinct from omniscience (all-knowing with infinite knowledge)
Rationality <> Perfection
‘Looking’ actions - information gathering, exploration
Learning
Not completely known environment
Improve, not all information from the designer
An agent is autonomous if its behavior is determined by its own experience (with ability to learn and adapt)
Performance measure: An objective criterion for success of an agent's behavior
E.g., performance measure of a vacuum-cleaner agent could be amount of dirt cleaned up with penalty for amount of time taken, amount of electricity consumed, amount of noise generated, etc.
Better: One point for a clean square at each time unit
Rational Agent: For each possible percept sequence, a rational agent should select an action that is expected to maximize its performance measure, given the evidence provided by the percept sequence and whatever built-in knowledge the agent has.
Rationality is distinct from omniscience (all-knowing with infinite knowledge)
Rationality <> Perfection
‘Looking’ actions - information gathering, exploration
Learning
Not completely known environment
Improve, not all information from the designer
An agent is autonomous if its behavior is determined by its own experience (with ability to learn and adapt)
Intelligent Agents
An agent is anything that can be viewed as perceiving its environment through sensors and acting upon that environment through actuators
Human agent:
Sensors: eyes, ears, and other organs
Actuators: hands, legs, mouth, and other body parts
Robotic agent:
Sensors: cameras, infrared range sensors, etc.
Actuators: various motors
Human agent:
Sensors: eyes, ears, and other organs
Actuators: hands, legs, mouth, and other body parts
Robotic agent:
Sensors: cameras, infrared range sensors, etc.
Actuators: various motors
Abridged history of AI
1943 McCulloch & Pitts: Boolean circuit model of brain based on artificial neurons
1950 Turing's "Computing Machinery and Intelligence“ – Turing test, subareas, machine learning, genetic algorithms
1952—69 Early enthusiasm, great expectations
1956 Dartmouth meeting: Phrase "Artificial Intelligence" adopted (John McCarthy, Minsky, Shannon, Newell/Simon…) Logic Theorist program
Next 20 years: dominated by MIT, CMU, Stanford, IBM
1950s Early AI programs, including Samuel's checkers program (learned to play better than its inventor), Newell & Simon's General Problem Solver, Gelernter's Geometry Engine
1958 Lisp McCarthy at MIT
1965 Robinson's complete algorithm for logical reasoning
Focus on Strong AI – solving math and logic problems, engaging in dialogues
Focus on Top-down approach – simulate concepts of human brain (planning, reasoning, language understanding, …)
Bottom-up approaches emerge – model low-level concepts (neurons, learning at much lower level, …)
Traditional top-down approach = Good-Old-Fashioned –AI (GOFAI)
Neat / Scruffy approaches
Neat: formal, pure, provable approach
Scruffy, messy: less provable but still yielding useful and significant results
AI’s Winter (a dose of reality) 1966-1973-mid 1980’s-now?
Predictions didn’t materialize
Simon 1957: in 10 years chess champion, major theorem proven. Actually, in 40 years.
Domain knowledge needed (eg automatic translation: ‘the spirit is willing but the flesh is weak’ <> ‘the vodka is good but the meat is rotten’ – US government funding on automatic translation projects cancelled.
AI discovers computational complexity – intractable problems.
Neural network research almost disappears (small mutation of programs …)
Negative results, eg. 1969 Minsky/Papert‘s Perceptrons
1969—79 Early development of knowledge-based systems
Early AI based on general-purpose search mechanisms. Weak methods – do not scale up. Alternative – use powerful domain-specific knowledge base
DENDRAL (first knowledge-intensive system)– infer molecular structure from info from mass spectrometer
Knowledge-based systems, expert systems; intensive work on knowledge representation
1973 Prolog (in France)
1980-present Results-oriented applications
Expert systems
NL systems
…
1981 Fifth Generation project (Japan), MCC (US) – never met their ambitious goals but lots of useful results, chips, …
1986– AI re-emerges
Strong AI – goal to emulate the full range of human cognitive capabilities
Weak AI – solve specific problems
Now focus on Weak AI, more realistic goals
Neural networks return to popularity
Connectionist models
Fuzzy logic, fuzzy controllers
Speech recognition and generation
Numerous applications
Artificial Life systems
Artificial Immune systems
Algorithms transition from AI algorithms to standard algorithms once they become practically useful
1987-- AI becomes a science
Build on existing theories rather than invent new
Real-world applications
A proper scientific methods – hypotheses are subjected to rigorous experiments, statistical analysis of results
Data mining
Gentle revolutions in robotics, computer vision, knowledge representation
1995-- The emergence of intelligent agents
Complete agent architecture
Web bots
1999 NASA – Deep Space I spacecraft – an agent to provide autonomy of spacecraft for limited durations of time
1950 Turing's "Computing Machinery and Intelligence“ – Turing test, subareas, machine learning, genetic algorithms
1952—69 Early enthusiasm, great expectations
1956 Dartmouth meeting: Phrase "Artificial Intelligence" adopted (John McCarthy, Minsky, Shannon, Newell/Simon…) Logic Theorist program
Next 20 years: dominated by MIT, CMU, Stanford, IBM
1950s Early AI programs, including Samuel's checkers program (learned to play better than its inventor), Newell & Simon's General Problem Solver, Gelernter's Geometry Engine
1958 Lisp McCarthy at MIT
1965 Robinson's complete algorithm for logical reasoning
Focus on Strong AI – solving math and logic problems, engaging in dialogues
Focus on Top-down approach – simulate concepts of human brain (planning, reasoning, language understanding, …)
Bottom-up approaches emerge – model low-level concepts (neurons, learning at much lower level, …)
Traditional top-down approach = Good-Old-Fashioned –AI (GOFAI)
Neat / Scruffy approaches
Neat: formal, pure, provable approach
Scruffy, messy: less provable but still yielding useful and significant results
AI’s Winter (a dose of reality) 1966-1973-mid 1980’s-now?
Predictions didn’t materialize
Simon 1957: in 10 years chess champion, major theorem proven. Actually, in 40 years.
Domain knowledge needed (eg automatic translation: ‘the spirit is willing but the flesh is weak’ <> ‘the vodka is good but the meat is rotten’ – US government funding on automatic translation projects cancelled.
AI discovers computational complexity – intractable problems.
Neural network research almost disappears (small mutation of programs …)
Negative results, eg. 1969 Minsky/Papert‘s Perceptrons
1969—79 Early development of knowledge-based systems
Early AI based on general-purpose search mechanisms. Weak methods – do not scale up. Alternative – use powerful domain-specific knowledge base
DENDRAL (first knowledge-intensive system)– infer molecular structure from info from mass spectrometer
Knowledge-based systems, expert systems; intensive work on knowledge representation
1973 Prolog (in France)
1980-present Results-oriented applications
Expert systems
NL systems
…
1981 Fifth Generation project (Japan), MCC (US) – never met their ambitious goals but lots of useful results, chips, …
1986– AI re-emerges
Strong AI – goal to emulate the full range of human cognitive capabilities
Weak AI – solve specific problems
Now focus on Weak AI, more realistic goals
Neural networks return to popularity
Connectionist models
Fuzzy logic, fuzzy controllers
Speech recognition and generation
Numerous applications
Artificial Life systems
Artificial Immune systems
Algorithms transition from AI algorithms to standard algorithms once they become practically useful
1987-- AI becomes a science
Build on existing theories rather than invent new
Real-world applications
A proper scientific methods – hypotheses are subjected to rigorous experiments, statistical analysis of results
Data mining
Gentle revolutions in robotics, computer vision, knowledge representation
1995-- The emergence of intelligent agents
Complete agent architecture
Web bots
1999 NASA – Deep Space I spacecraft – an agent to provide autonomy of spacecraft for limited durations of time
Acting rationally: rational agent
Rational behavior: doing the right thing
The right thing: that which is expected to maximize goal achievement, given the available information
Doesn't necessarily involve thinking – e.g., blinking reflex – but thinking should be in the service of rational action
An agent is an entity that perceives and acts
Intelligent agents – more than mere programs – attributes such as operating autonomously, perceiving their environment, adapting to change
Uncertainty may be involved
For any given class of environments and tasks, we seek the agent (or class of agents) with the best performance
Computational limitations make perfect rationality unachievable
Logic reasoning is needed but not enough
Correct inference is just one of several possible mechanisms for achieving rationality
All the skills from the Turing test are needed
The right thing: that which is expected to maximize goal achievement, given the available information
Doesn't necessarily involve thinking – e.g., blinking reflex – but thinking should be in the service of rational action
An agent is an entity that perceives and acts
Intelligent agents – more than mere programs – attributes such as operating autonomously, perceiving their environment, adapting to change
Uncertainty may be involved
For any given class of environments and tasks, we seek the agent (or class of agents) with the best performance
Computational limitations make perfect rationality unachievable
Logic reasoning is needed but not enough
Correct inference is just one of several possible mechanisms for achieving rationality
All the skills from the Turing test are needed
Subscribe to:
Posts (Atom)