Introduction to Intelligent Systems PDF

Introduction to Intelligent Systems Ezequiel López-Rubio and Enrique Domínguez Department of Computer Languages and Computer Science University of Málaga, Spain Contents  1. Artificial Intelligence (AI)  2. History of AI  3. Intelligent agents What is artificial intelligence? 1. ARTIFICI...

Introduction to Intelligent Systems Ezequiel López-Rubio and Enrique Domínguez Department of Computer Languages and Computer Science University of Málaga, Spain Contents  1. Artificial Intelligence (AI)  2. History of AI  3. Intelligent agents What is artificial intelligence? 1. ARTIFICIAL INTELLIGENCE 1. Artificial Intelligence (AI) The art of creating machines that Definitions perform functions that require intelligence when performed by people  Acting humanly Kurzweil, 1990 ◼ The Turing Test The automation of activities that we associate with human thinking,  Thinking humanly activities such that decision-making, problem solving, learning… ◼ Cognitive science Bellman, 1978 ◼ Neuroscience The study of computations that make it possible to perceive,  Thinking rationally reason, and act Aristotle’s syllogisms Winston, 1992 ◼  Acting rationally AI is concerned with intelligent behavior in artifacts ◼ Rational agents Nilsson, 1998 1. Artificial Intelligence (AI) Acting humanly  Can machines think? ◼ Can they exhibit an intelligent behavior?  Turing Test NON-reproducible NON-constructive ◼ Process natural language NO mathematical analysis ◼ Represent knowledge ◼ Reason ◼ Learn ◼ Perceive objects ◼ Manipulate objects 1. Artificial Intelligence (AI) Thinking humanly  How does a human think? ◼ How does the brain work?  Validation of theories of human mind ◼ Prediction and comparison with human behavior (Cognitive Science) Cognitive ◼ Identification with neural behavior and Science implementation (Neuroscience) Neuroscience AI 1. Artificial Intelligence (AI) Thinking rationally  Aristotle’s syllogisms Socrates is a man ◼ Codify ‘right thinking’ All men are mortal Ⱶ  Logic Socrates is mortal ◼ Notation ◼ Inference rules ◼ An automated process? Mathematics Modern Logic AI Philosophy 1. Artificial Intelligence (AI) Acting rationally  Rational behavior ◼ Act so as to achieve the best expected outcome  Achieve one’s goals efficiently in a given environment  It is not always necessary to think/reason  Agent ◼ Entity whose actions are based on the perceived information 2. HISTORY OF AI 2. History of AI The inception of AI (1943-1956)  W. McCulloch & W. Pitts (1943) ◼ Hebbian learning (1949) ◼ SNARC – First neural network computer (1950)  A. Turing ‘Computing Machinery and Intelligence’ (1950) ◼ Can machines think?  E. Dijkstra: Can submarines swim? ◼ Turing Test 2. History of AI Early enthusiasm (1952-1969)  First programs ◼ Nevell & Simon – General Problem Solver ◼ A. Samuel – Checkers Program ◼ H. Gelernter – Geometry Theorem Prover  Dartmouth workshop (1956) ◼ 10 people during 2 months “Look, Ma,  Lisp (John McCarthy, 1958) no hands!”  Genetic algorithms (1958-59) ◼ Friedberg – Machine evolution  Neural networks (1962) ◼ Widrow – Adalines ◼ Rosenblatt – Perceptron 2. History of AI Expert systems (1969-1986)  “A dose of reality” (1966-1973) ◼ Overconfidence on the performance of AI systems for more complex problems ◼ Illusion of unlimited computational power ◼ Criticisms (Lighthtill report, 1973) and lack of funding (“The AI winter”) ◼ Limitations of basic structures  Computational complexity ◼ General purpose programs (weak methods) → Domain specific programs  Expert systems (1969) ◼ DENDRAL, MYCIN,… ◼ Prolog  RI – First commercial expert system (1982) ◼ Digital Equipment Corporation (McDermott) ◼ Saved more than $40 million / year 2. History of AI Probabilistic reasoning and ML (80-90s)  The return of neural networks: Back-propagation algorithm ◼ Rumelhart & McClelland (1986) – Parallel Distributed Processing  Probability in IA ◼ More scientific approach ◼ Benchmark datasets and competitions to demonstrate progress ◼ Hidden Markov models: mathematical theory + training ◼ Bayesian networks: uncertain knowledge representation + algorithms for reasoning (‘Probabilistic Reasoning in Intelligent Systems’, J. Pearl, 1988)  Reunification of subfields ◼ Computer vision, speech recognition, natural language processing… ◼ New applications and faster deployment (robotics) ◼ Better understanding of the core problems in AI 2. History of AI Big Data and Deep Learning (2000s)  Appearance of very large datasets thanks to the World Wide Web ◼ Unstructured data: text, images, audio, video, etc. ◼ New algorithms to deal with unlabeled data ◼ AI starts attracting commercial attention (Watson systems – IBM)  Deep learning (2011-) ◼ Multiple layers of simple, adjustable computing elements ◼ Convolutional neural networks (1990s) found new applications in speech recognition and visual object recognition ◼ More applications emerge: machine translation, medical diagnosis, game playing ◼ DL relies heavily on powerful hardware (GPU) ◼ Resurgence of interest in AI  The present (and future?) ◼ Next generation of conversational assistants (Chatbots) ◼ Generative AI (multimodal) 3. INTELLIGENT AGENTS 3. Intelligent agents Definition  Agent: entity which acts based on its perception of its environment ◼ Environment Environment ◼ Sensors Perception ◼ Actuators Action  f : P* → A Agent  Agent = Program + Architecture 3. Intelligent agents Definition  Intelligent agent: According to the received percepts, it selects the right actions to optimize its expected performance (so as to attain its goal) ◼ Do the right thing ◼ Goals are desirable states of the environment ◼ Evaluation of the agent’s performance 3. Intelligent agents Omniscience, learning and autonomy  Intelligent ≠ Omniscient ◼ Not all the relevant information can be perceived  Information gathering (exploration) ◼ The outcomes of actions might not be as expected  Autonomous (learning) ◼ Behavior determined by its own experience 3. Intelligent agents Specification  Performance Evaluation  Environment  Actuators  Sensors 3. Intelligent agents Properties of task environments  Observable: partially / fully  Single agent / multiagent ◼ Competitive / Cooperative  Deterministic / stochastic  Episodic / sequential  Static / dynamic  Discrete / continuous  Known / unknown 3. Intelligent agents Structure Artificial intelligence Find the way to implement f : P* → A Architecture Program Agent 3. Intelligent agents Kinds of intelligent agents  Rule-based (reactive) ◼ Actions depend on current percepts, ignoring the rest of the percept history (if-then)  Model-based (with internal states) ◼ The current state depends on the percept history  Goal-based ◼ Search and planning  Utility-based ◼ Maximize the expected utility of action outcomes 3. Intelligent agents Reactive Percept Environment Sensors Information gathering Action Action Rules selection Agent Actuators 3. Intelligent agents Model-based Percept Environment Sensors Model Information gathering Action Action Rules selection Agent Actuators 3. Intelligent agents Goal-based Percept Environment Sensors Model Information gathering Action Action Goals selection Agent Actuators 3. Intelligent agents Utility-based Percept Environment Sensors Model Information gathering Action Action Utilities selection Agent Actuators 3. Intelligent agents Learning agents Performance standard Percept Environment Sensors Critic Information gathering Learning Action Action selection Problem generator Agent Actuators