Publicité
Publicité

Contenu connexe

Publicité

Introduction to artificial intelligence

  1. Introduction to Artificial Intelligence (AI) Dr. Albert Fornells Herrera http://www.linkedin.com/in/afornells
  2. 1. To have or not to have mind…this is the question 2. Why AI is interesting for industry? 3. What does “intelligence” mean? 4. What is Artificial Intelligence? 5. Types of AI 6. Why is AI powerful 7. Bases of AI 8. Artificial intelligence history: From 'dark ages' to KBS 9. Artificial intelligence map 10. References 2 Outline Introduction to Artificial Intelligence - Albert Fornells
  3.  Philosophers have been trying for over two thousand years to understand and resolve two big questions of the universe:  how does a human mind work, and  can non-humans have minds?  However, these questions are still unanswered  Some philosophers have picked up the computational approach originated by computer scientists and accepted the idea that machines can do everything that humans can do  Others have an opposed idea, claiming that such highly sophisticated behavior as love, creative discovery and moral choice will always be beyond the scope of any machine  Engineers and scientists have already built machines that can be called ‘intelligent’.  So what does the word ‘intelligence’ mean? And intelligent machine? 3 To have or not to have mind…this is the question Introduction to Artificial Intelligence - Albert Fornells
  4.  Main reasons:  Economical needs  Experts highly qualified are expensive and they cannot afford them  A way for training new experts  Knowledge preservation  Computational efficiency  General decision methods are slowly and inefficient  Massive information needs to be processed  Decision support system  Get fast, efficient and justified decisions  Autonomous decisions systems  Summarizing: Companies want to build specific systems to solve specific problems using the knowledge from domain and the expertise from experts  They want to introduce our “minds” and “human capabilities” in their systems 4 Why AI is interesting for industry? Introduction to Artificial Intelligence - Albert Fornells
  5.  Let us look at a dictionary definition (Essential English Dictionary,Collins, 1990)  Def 1. Someone’s intelligence is their ability to understand and learn things  Def 2. Intelligence is the ability to think and understand instead of doing things by instinct or automatically  So, it can be defined as ‘the ability to learn and understand, to solve problems and to make decisions’  Building a machine that is (or seems to be) intelligent is what is called intelligent machine.  However, the intelligence is not the same for everyone (and also for machines), 5 What does “intelligence” mean? Introduction to Artificial Intelligence - Albert Fornells is not the same as
  6.  John McCarthy coined AI term in 1956 as ‘the science and engineering of making intelligent machines’ at a conference at Dartmouth College. Intelligent machine terms refer to the capability of performing intelligent human processes as:  Learning  Reasoning  Problem solving  Perception  Language understanding  AI has become an essential part of the technology industry, providing the heavy lifting for many of the most difficult problems in computer science.  Prediction  Classification  Regression  Clustering  Function optimization 6 What is Artificial Intelligence (AI)? Introduction to Artificial Intelligence - Albert Fornells
  7. 7 So, an intelligent machine is a… Introduction to Artificial Intelligence - Albert Fornells System that acts… System that thinks… …like humans ...rationally Make tasks that, nowadays, are better done by humans (Rich y Knight, 1991) Exploring and emulating the intelligent behavior in computational process terms (Shalkoff, 1990) Study of mental capabilities through the study of computational models (Charniak y McDermott, 1985) Computer with minds at literal sense (Haugeland, 1985)
  8.  The goal is to build a system that seems to be a human  At 1950, Alain Turing proposed the Turing test in the article “Computing machinery and intelligence” in the Mind Journal  Instead of asking, ‘Can machines think?’, Turing said we should ask, ‘Can machines pass a behavior test for intelligence?’  So, Turing defined the intelligent behavior of a computer as the ability to achieve the human-level performance in cognitive tasks  A person -acting as a judge- has a conversation for 5 minutes through a text channel. The judge has to identify if is talking with a computer or a human  Required capabilities:  Natural language processing  Knowledge representation  Reasoning  Learning 8 Systems that act like humans Introduction to Artificial Intelligence - Albert Fornells
  9.  Systems need to act as a human.  They have to lie and make mistakes as us!!  Loebner prize: Turing test competition.  Total Turing test: The same but also with signal video. 9 More about Turing test Introduction to Artificial Intelligence - Albert Fornells [INT] I heard that a striped rhinoceros flow on the Mississippi in a pink balloon this morning. What do you think about? [COMP] That sound rather ridiculous to me [INT] Really? My uncle did this one... Why this sound ridiculous? [COMP] Option 1: Rhinoceros don't have stripes [COMP] Option 2: Rhinoceros can't fly [INT] What’s the result of 324 x 678? [COMP] This is too difficult. I’m not a calculator!
  10.  The goal is to model the human brain and set a theory about how it works.  Theory should allow the definition of computational models.  This approach is highly influenced by neuroscience and cognitive sciences.  The conscious mystery  Many religions consider that consciousness resides in the soul.  Can it be included in machines?  D. Chalmers tackles the conscious topic from two points of view:  Easy: Distinguish between conscious and unconscious thinking (Freud).  Heart beat control, muscular movements, planning a day, etc.  Difficult: Explain how the subjective experience is born from neurons.  Surprising hypothesis from Francis Crick (1916-2004): brain as a machine  Our thoughts, sensations and feelings are the result of the physiological activity of the brain tissues.  Conscious is a biological product such as blood circulation or the digestion. 10 Systems that think like humans Introduction to Artificial Intelligence - Albert Fornells
  11.  Aristotle (384 BC- 322 BC) was the first to codify “the right thinking”, that is, irrefutable reasoning process.  Rationally thoughts are based on the logic.  Formal Logic can be introduced and applied in computers.  But….  It is complex formalize the knowledge  There is a big difference between being able to solve a problem “in principle” and doing so in practice. 11 Systems that think rationally Introduction to Artificial Intelligence - Albert Fornells
  12.  To act rationally means to achieve a set of goals using a previous knowledge base.  Approach focuses on rational agents.  An agent receives and acts taking into account the environment.  The aim is not to imitate the human model, but to complete a set of actions to achieve a final goal.  An agent needs:  Perception  Natural language processing  Knowledge representation  Reasoning  Machine learning 12 Systems that act rationally Introduction to Artificial Intelligence - Albert Fornells
  13.  The power resides in the combination of disciplines that tackle the same problems as AI: learn and understand, to solve problems and to make decisions.  AI is fed from many disciplines  Philosophy: Logic, methods of reasoning, mind as physical system, foundations of learning, language, rationality.  Mathematics: Formal representation and proof, algorithms, computation, (un)decidability, (in)tractability.  Statistics : Modeling uncertainty, learning from data.  Economics: Utility, decision theory, rational economic agents.  Neuroscience: Neurons as information processing units.  Psychology / NeuroScience: How do people behave, perceive, process cognitive information, represent knowledge.  Computer Engineering: Building fast computers.  Control Theory: Design systems that maximize an objective function over time.  Linguistics: Knowledge representation, grammars. 13 Why AI is powerful? Introduction to Artificial Intelligence - Albert Fornells
  14. 14 Bases of AI Introduction to Artificial Intelligence - Albert Fornells Philosophy • Discussion about the possibility of a mechanical intelligence. Mathematical. • Philosophic bases requires formal rules. Computational linguistic • Understanding language requires understanding of the subject matter and the context. Cognitive psychology • Behavior theories, rational behave bases. Computational engineering • Some mechanism, hardware and tools are required for AI.
  15.  Philosophy: Discussion about the possibility of a mechanical intelligence.  Descartes (1556-1650), Leibniz (1646-1716): mind is linked to the physical world.  John Locke: In the beginning there was a Mind (1960).  Hume (1711-1776), Russell (1872-1970): the knowledge comes from the perception, its acquired by the experience (induction) and its represented by the logical theories.  Darwin (1809-1882): Evolutionary theory by natural selection at 1859.  Mathematical. Philosophic bases require formal rules.  Boole (1815-1864), Frege (1848-1925): Logical mathematical fundamentals.  Gödel (1906–1978), Turing (1912–1954): Computability limits.  Fermat (1601–1665), Bernoulli (1700–1782), Bayes (1702-1761): Probability and Probabilistic reasoning. 15 Bases of AI Introduction to Artificial Intelligence - Albert Fornells
  16.  Cognitive psychology. Behavior theories, rational behave bases.  William James (1842-1910): Brain possess and processes information.  Hermann Von Helmholtz (1821-1894): Perception involves a form of unconscious logical inference.  Kenneth Craik (1914-1945): Put back the missing mental step between stimulus and response.  Stimulus must be translated into an internal representation.  The representation is manipulated by cognitive process to derive a new internal representation.  These are in turn retranslated back into actions.  Computational linguistic: Understanding language requires understanding of the subject matter and the context.  B.F. Skinner (1904-1990) published Verbal Behavior at 1957 about the language learning.  Noam Chomsky (1928-present) published Syntactic Structures at 1957 about knowledge representation and language grammar. 16 Bases of AI Introduction to Artificial Intelligence - Albert Fornells
  17.  Computational engineering: Some mechanisms, hardware and tools are required for AI  Heath Robinson (1940), first operational modern computing for the single purpose of deciphering German messages built by Alain Turing  Z3 (1943), first operational programmable computer built by Konrad Zuse  ENIAC (1945), first general-purpose, electronic and digital computer built by John Machly and John Eckert  EDVAC (1949), followed John Von Neumann’s suggestion to use a stored program  6000 wave tube, 12000 diodes, 56kw, 45’5m2 and 7850kg. Need 30 workers 17 Bases of AI Introduction to Artificial Intelligence - Albert Fornells ENIAC EDVAC Microprocessor
  18. 18 AI history: From ‘Dark ages’ to KBS Introduction to Artificial Intelligence - Albert Fornells early 1990s-onwards The failure of expert systems and where AI is going mid 1980s-early 1990s AI becomes an industry early 1970s–mid-1980s The technology of expert systems late 1960s–early 1970s Unfulfilled promises 1956–late 1960s The era of great expectations 1943–1956 The birth of artificial intelligence
  19.  Situation: There was nothing  First neuron model and demonstration that any function is computable (Warren McCulloch and Walter Pits, 1943)  Hebbian theory describes a basic mechanism for synaptic plasticity (Donald Hebb, 1949)  First neural network simulator (Marvin Misky and Dean Edmonds, 1951)  ENIAC (1946) and EDVAC (1949) are built  Demonstrating the need to use heuristics in the search for the chess game (Claude Shannon, 1950)  Creation of the non numerical reasoning program “Logic Theorist” (Herbert Newell, Allen Simon, J.C. Shaw, 1955)  Summer Workshop at Dartmouth College where AI concept is founded (McCarthy, Minsky, Shannon, Rochester, More, Samuel, Solomonoff, Selfridge, Newell and Simon, 1956) 19 AI history: Beginnings (1943-1956) Introduction to Artificial Intelligence - Albert Fornells
  20.  The early years of AI are characterized by tremendous enthusiasm, great ideas and very limited success.  Only a few years before, computers had been introduced to perform routine mathematical calculations.  Now AI researchers were demonstrating that computers could do more than that. It was an era of great expectations.  Definition of the high-level language LISP (John McCarthy, 1958).  Advice Taker is proposed to search for solutions to general problems of the world. Focus on formal logic (John McCarthy, 1958).  Generate solutions using axioms and new ones can be added.  1st KBS with the principles of knowledge representation and reasoning.  Frame theory for knowledge representation and reasoning (Marvin Minsky, 1975).  Perceptron convergence theorem is proved (Frank Rosenblatt, 1962). 20 AI history: Big expectations (1956-late 1960s) Introduction to Artificial Intelligence - Albert Fornells
  21.  Development GPS, a General Problem Solver, to simulate human problem- solving methods (Herbert Newell and Allen Simon,1961)  First attempt to separate the problem solving-technique from the data  Means-ends analysis: Strategy based on states  GPS fails in complicated problems due to the infinite number of states  Fuzzy logic is defined (Lofti Zadeh, 1965)  By 1980 the euphoria about AI was gone and most of government funding for AI projects was cancelled  AI researches attempted to simulate the complex thinking process by inventing general methods for solving broad classes of problems.  The general purpose search mechanism not work with real problems. They were called weak methods because they use weak information from domain  However, it was a time when new ideas such as knowledge representation, learning and computing concepts were set 21 AI history: Big expectations (1956-late 1960s) Introduction to Artificial Intelligence - Albert Fornells
  22.  From the mid-1950s, AI researchers were making promises to build all-purpose intelligent machines on a human-scale knowledge base by the 1980s and to exceed human intelligence by the year 2000.  By 1970, however, they realized that such claims were too optimistic.  Few AI programs could demonstrate some level of machine intelligence in one or two toy problems.  Main difficulties:  General methods for broad classes of problems. So, NO knowledge about domain  Search strategies based on try and error. They were not scaled for big problems.  Easy or tractable problems can be solved in polynomial time, intractable problems require times that are exponential functions of the problem size.  So, many problems were NP-hard. “The size matters”  Eliza, the first chatterbot is created (Joseph Weizenbaum, 1966).  Probably the most important development in the 1970s was the realization that the problem domain for intelligent machines had to be sufficiently restricted 22 AI history: A dose or reality (late 60s-early 70s) Introduction to Artificial Intelligence - Albert Fornells
  23.  Knowledge from expert is used to tackle real problems  DENDRAL: A molecular structure interpretation (Feigenbaum,1971)  The first expert system to solve a real problem  The major shift in AI: shift from general purpose, knowledge-sparse, weak methods to domain-specific, knowledge intensive techniques  A new methodology of expert systems: knowledge engineering, which encompassed techniques of capturing, analyzing and expressing in rules an expert’s ‘know-how’. Appears the knowledge acquisition bottleneck problem  MYCIN: Rule-based expert system for the diagnosis of infectious blood diseases (Edgar ShortLiffe, 1972)  Worked as well as experts and better than junior doctors  Which were the main contributions?  Knowledge in the form of rules was clearly separated from the reasoning mechanism  No general theoretical model existed: Rules are extracted from interviews  Rules reflect the uncertainty with medical knowledge 23 AI history: Knowledge as key of success (early 70s- mid 80s) Introduction to Artificial Intelligence - Albert Fornells
  24.  PROSPECTOR . A computer-based consultation system for mineral exploration (Duda, 1981)  It used a combined structure that incorporated rules and a semantic network  It had a sophisticated support package including a knowledge acquisition system  It incorporated Bayes’ rules of evidence to propagate uncertainties through the system  Many applications showed that AI technology could move successfully from the research laboratory to the commercial environment  Most expert systems were developed with special AI languages, such as LISP, PROLOG and OPS, based on powerful workstations  Expensive hardware and complicated programming language means few researchers  Only in 1980, with the arrival of PCs and easy-to-use expert system development tools made possible the expansion to more modest laboratories  Knowledge representation and reasoning were the key for success 24 AI history: Knowledge as key of success (early 70s- mid 80s) Introduction to Artificial Intelligence - Albert Fornells
  25.  A survey found nearly 200 expert systems, most of them applications in the field of medical diagnosis (Waterman, 1986)  7 years later, the survey found 2500 developed expert systems (Durkin, 1994). The new growing area was business and manufacturing  Expert system technology had clearly matured, but are the key of success in any field? It would be a mistake to overestimate the capability of this technology  Expert systems are restricted to a very narrow domain of expertise  Expert systems are not as robust and flexible as a user might want  Expert systems have limited explanation capabilities  Expert systems are also difficult to verify and validate  Expert systems, especially the first generation, have little or no ability to learn from their experience  Despite all these difficulties, expert systems have made the breakthrough and proved their value in a number of important applications 25 AI history: AI becomes an industry (mid 80s-begin90s) Introduction to Artificial Intelligence - Albert Fornells
  26.  Disillusion about the applicability of expert system technology  Building an expert system required much more than just buying a reasoning system or expert system shell and putting enough rules in it  Expert systems were too expensive to maintain  They were difficult to update, they could not learn, they were "brittle“ (i.e. they could make grotesque mistakes when given unusual inputs)  They proved useful, but only in a few special contexts  So, the next step was to evolve the Expert System to Knowledge Base Systems 26 AI history: The failure of expert systems (begin 90s – onwards) Introduction to Artificial Intelligence - Albert Fornells Emulate the resolution capability of an expert human  Building through knowledge engineering Mainly based on rules Reduced learning capabilities Use the knowledge from domain to solve problems Include automatic acquisition process inside the knowledge engineering process  Heterogeneous methodologies and architectures (rules, cases, models, etc.) Extended learning capabilities
  27. 27 A possible map of the current AI Introduction to Artificial Intelligence - Albert Fornells • Robotics and control • Natural Language Processing • Artificial vision • Speech recognition • Logic • Multiagents systems • Decision Support System • Knowledge management • Knowledge representation • Ontology and semantic web • Computer-Human interaction • Evolutionary Computation • Case-Based Reasoning • Reinforcement Learning • Neural Network • Data Analysis • Non monotonic reasoning • Model based reasoning • Constraint satisfaction • Qualitative reasoning • Uncertain reasoning • Temporal reasoning • Heuristic search Reasoning Machine Learning Robotics, perception and natural language processing Knowledge Management
  28. 1. For each of the following, give reasons why: a) A dog is more intelligent than a worm b) A human is more intelligent than a dog c) An organization is more intelligent than an individual human Based on these, give a definition of what “more intelligent” may mean 2. Choose a particular word, for example, what is on some part of your desk at the current time i. Get someone to list all of the things that exist in this world (or try it yourself as a thought experiment) ii. Try to think of ten things that they missed. Make these as different from each other as possible iii. Try to find a thing that can’t be described using natural language iv. Choose a particular task, such as making the desk tidy, and try to write down all of the things in the world at a level of description that is relevant to this task 28 Exercises Introduction to Artificial Intelligence - Albert Fornells
  29. 3. Look for the “Chinese room argument” significance (Searle, 1980). Which was its goal? 29 Exercises Introduction to Artificial Intelligence - Albert Fornells
  30.  S. Russell, P. Norvig. Artificial Intelligence: a Modern Approach, 3rd ed. Pearson Higher Education, 2010 (*).  Also available in Spanish.  M. Negnevitsky, Artificial Intelligence, A Guide to Intelligent Systems. Addison Wesley, 2002 (*).  N. J. Nilsson, Artificial Intelligence: a new synthesis, The Morgan Kaufmann Series in Artificial Intelligence Series. Morgan Kaufmann, 1998 (*). (*) Partially published in Google Books 30 References Introduction to Artificial Intelligence - Albert Fornells
Publicité