Artificial intelligence

From Wikipedia, the free encyclopedia.

This article is about intelligence exhibited by manufactured systems, typically computers. For other uses of the term AI, see Ai.

Artificial intelligence (also known as machine intelligence and often abbreviated as AI) is intelligence exhibited by any manufactured (i.e. artificial) system. The term is often applied to general purpose computers, and also in the field of scientific investigation into the theory and practical application of AI.

Artificial intelligence can be considered in two parts: "What is the nature of artifice?" and "What is intelligence?" The first question is relatively easy, although it leads to an examination what can be manufactured. The limitations of classical computational systems, available manufacturing processes, or human intellect may all place constraints on what can be manufactured.

The second question raises ontological issues of consciousness and intelligence as displayed by humans, as intelligent behavior in humans is complex and often difficult to understand. Study of animals and artificial systems which are not simply models of what already exists are also considered highly relevant.

Strong AI and weak AI

One popular and early definition of artificial intelligence research, put forth by John McCarthy at the Dartmouth Conference in 1956, is "making a machine behave in ways that would be called intelligent if a human were so behaving", repeating the claim put forth by Alan Turing in "Computing machinery and intelligence" (Mind, October 1950). However this definition seems to ignore the possibility of strong AI (see below). Another definition of artificial intelligence is "intelligence arising from an artificial device". Most definitions could be categorized as concerning either systems that think like humans, systems that act like humans, systems that think rationally or systems that act rationally.

Strong artificial intelligence

Strong AI research deals with the creation of AI that can truly reason and solve problems; a strong AI is said to be sentient, or self-aware, but may or may not exhibit human-like thought processes.

Weak artificial intelligence

Weak artificial intelligence research deals with the creation of some form of computer-based artificial intelligence that can reason and solve problems only in a limited domain; such a machine would, in some ways, act as if it were intelligent, but it would not possess true intelligence or sentience. The classical test for such abilities is the Turing test.

There are several fields of weak AI, one of which is natural language. Many weak AI fields have specialised software or programming languages created for them. For example, the 'most-human' natural language chatterbot A.L.I.C.E. uses a programming language AIML that is specific to its program, and the various clones, named Alicebots. Jabberwacky is a little closer to strong AI, since it learns how to converse from the ground up based solely on user interactions.

To date, much of the work in this field has been done with computer simulations of intelligence based on predefined sets of rules. Very little progress has been made in strong AI. Depending on how one defines one's goals, a moderate amount of progress has been made in weak AI.

When viewed with a moderate dose of cynicism, weak artificial intelligence can be viewed as ‘the set of computer science problems without good solutions at this point.’ Once a sub-discipline results in useful work, it is carved out of artificial intelligence and given its own name. Examples of this are pattern recognition, image processing, neural networks, natural language processing, robotics and game theory. While the roots of each of these disciplines is firmly established as having been part of artificial intelligence, they are now thought of as somewhat separate.

Philosophical criticism and support of strong AI

The term "Strong AI" was originally coined by John Searle and was applied to digital computers and other information processing machines. Searle defined strong AI:

"according to strong AI, the computer is not merely a tool in the study of the mind; rather, the appropriately programmed computer really is a mind" (J Searle in Minds Brains and Programs. The Behavioral and Brain Sciences, vol. 3, 1980).

Searle and most others involved in this debate are addressing the problem of whether a machine that works solely through the transformation of encoded data could be a mind, not the wider issue of Monism versus Dualism (ie: whether a machine of any type, including biological machines, could contain a mind).

Searle states in his Chinese Room argument that information processors carry encoded data which describe other things. The encoded data itself is meaningless without a cross reference to the things it describes. This leads Searle to point out that there is no meaning or understanding in an information processor itself. As a result Searle claims to demonstrate that even a machine that passed the Turing test would not necessarily be conscious in the human sense.

Some philosophers hold that if Weak AI is accepted as possible then Strong AI must also be possible. Daniel C. Dennett argues in Consciousness Explained that if there is no magic spark or soul, then Man is just a machine, and he asks why the Man-machine should have a privileged position over all other possible machines when it comes to intelligence or 'mind'. Simon Blackburn in his introduction to philosophy, Think, points out that you might appear intelligent but there is no way of telling if that intelligence is real (ie: a 'mind'). However, if the discussion is limited to strong AI rather than artificial consciousness it may be possible to identify features of human minds that do not occur in information processing computers.

Strong AI seems to involve the following assumptions about the mind and brain:

the mind is software, a finite state machine so the Church-Turing thesis applies to it
the brain is purely hardware (i.e. only follows the rules of a classical computer)

The first assumption is particularly problematic because of the old adage that any computer is just a glorified abacus. It is indeed possible to construct any type of information processor out of balls and wood, although such a device would be very slow and prone to failure, it would be able to do anything that a modern computer can do. This means that the proposition that information processors can be minds is equivalent to proposing that minds can exist as devices made of rolling balls in wooden channels.

Some (including Roger Penrose) attack the applicability of the Church-Turing thesis directly by drawing attention to the halting problem in which certain types of computation cannot be performed by information systems yet seem to be performed by human minds.

Ultimately the truth of Strong AI depends upon whether information processing machines can include all the properties of minds such as Consciousness. However, Weak AI is independent of the Strong AI problem and there can be no doubt that many of the features of modern computers such as multiplication or database searching might have been considered 'intelligent' only a century ago.

History

Development of AI theory

Much of the (original) focus of artificial intelligence research draws from an experimental approach to psychology, and emphasizes what may be called linguistic intelligence (best exemplified in the Turing test).

Approaches to artificial intelligence that do not focus on linguistic intelligence include robotics and collective intelligence approaches, which focus on active manipulation of an environment, or consensus decision making, and draw from biology and political science when seeking models of how "intelligent" behavior is organized.

Artificial intelligence theory also draws from animal studies, in particular with insects, which are easier to emulate as robots (see artificial life), as well as animals with more complex cognition, including apes, who resemble humans in many ways but have less developed capacities for planning and cognition. AI researchers argue that animals, which are simpler than humans, ought to be considerably easier to mimic. But satisfactory computational models for animal intelligence are not available.

Seminal papers advancing the concept of machine intelligence include A Logical Calculus of the Ideas Immanent in Nervous Activity (1943), by Warren McCulloch and Walter Pitts, and On Computing Machinery and Intelligence (1950), by Alan Turing, and Man-Computer Symbiosis by J.C.R. Licklider. See cybernetics and Turing test for further discussion.

There were also early papers which denied the possibility of machine intelligence on logical or philosophical grounds such as Minds, Machines and Gödel (1961) by John Lucas [1] (http://users.ox.ac.uk/~jrlucas/Godel/mmg.html).

With the development of practical techniques based on AI research, advocates of AI have argued that opponents of AI have repeatedly changed their position on tasks such as computer chess or speech recognition that were previously regarded as "intelligent" in order to deny the accomplishments of AI. They point out that this moving of the goalposts effectively defines "intelligence" as "whatever humans can do that machines cannot".

John von Neumann (quoted by E.T. Jaynes) anticipated this in 1948 by saying, in response to a comment at a lecture that it was impossible for a machine to think: "You insist that there is something a machine cannot do. If you will tell me precisely what it is that a machine cannot do, then I can always make a machine which will do just that!". Von Neumann was presumably alluding to the Church-Turing thesis which states that any effective procedure can be simulated by a (generalized) computer.

In 1969 McCarthy and Hayes started the discussion about the frame problem with their essay, "Some Philosophical Problems from the Standpoint of Artificial Intelligence".

Experimental AI research

Artificial intelligence began as an experimental field in the 1950s with such pioneers as Allen Newell and Herbert Simon, who founded the first artificial intelligence laboratory at Carnegie-Mellon University, and McCarthy and Marvin Minsky, who founded the MIT AI Lab in 1959. They all attended the aforementioned Dartmouth College summer AI conference in 1956, which was organized by McCarthy, Minsky, Nathan Rochester of IBM and Claude Shannon.

Historically, there are two broad styles of AI research - the "neats" and "scruffies". "Neat", classical or symbolic AI research, in general, involves symbolic manipulation of abstract concepts, and is the methodology used in most expert systems. Parallel to this are the "scruffy", or "connectionist", approaches, of which neural networks are the best-known example, which try to "evolve" intelligence through building systems and then improving them through some automatic process rather than systematically designing something to complete the task. Both approaches appeared very early in AI history. Throughout the 1960s and 1970s scruffy approaches were pushed to the background, but interest was regained in the 1980s when the limitations of the "neat" approaches of the time became clearer. However, it has become clear that contemporary methods using both broad approaches have severe limitations.

Artificial intelligence research was very heavily funded in the 1980s by the Defense Advanced Research Projects Agency in the United States and by the fifth generation computer systems project in Japan. The failure of the work funded at the time to produce immediate results, despite the grandiose promises of some AI practitioners, led to correspondingly large cutbacks in funding by government agencies in the late 1980s, leading to a general downturn in activity in the field known as AI winter. Over the following decade, many AI researchers moved into related areas with more modest goals such as machine learning, robotics, and computer vision, though research in pure AI continued at reduced levels.

Practical applications of AI techniques

Whilst progress towards the ultimate goal of human-like intelligence has been slow, many spinoffs have come in the process. Notable examples include the languages LISP and Prolog, which were invented for AI research but are now used for non-AI tasks. Hacker culture first sprang from AI laboratories, in particular the MIT AI Lab, home at various times to such luminaries as McCarthy, Minsky, Seymour Papert (who developed Logo there), Terry Winograd (who abandoned AI after developing SHRDLU).

Many other useful systems have been built using technologies that at least once were active areas of AI research. Some examples include:

Chinook was declared the Man-Machine World Champion in checkers (draughts) in 1994.
Deep Blue, a chess-playing computer, beat Garry Kasparov in a famous match in 1997.
InfoTame, a text analysis search engine developed by the KGB for automatically sorting millions of pages of communications intercepts.
Fuzzy logic, a technique for reasoning under uncertainty, has been widely used in industrial control systems.
Expert systems are being used to some extent industrially.
Machine translation systems such as SYSTRAN are widely used, although results are not yet comparable with human translators.
Neural networks have been used for a wide variety of tasks, from intrusion detection systems to computer games.
Optical character recognition systems can translate arbitrary typewritten European script into text.
Handwriting recognition is used in millions of personal digital assistants.
Speech recognition is commercially available and is widely deployed.
Computer algebra systems, such as Mathematica and Macsyma, are commonplace.
Machine vision systems are used in many industrial applications ranging from hardware verification to security systems.
AI Planning methods were used to automatically plan the deployment of US forces during Gulf War I. This task would have cost months of time and millions of dollars to perform manually, and DARPA stated that the money saved on this single application was more than their total expenditure on AI research over the last 50 years.

The vision of artificial intelligence replacing human professional judgment has arisen many times in the history of the field, in science fiction and today in some specialized areas where "expert systems" are used to augment or to replace professional judgment in some areas of engineering and of medicine.

Hypothetical consequences of AI

Some observers foresee the development of systems that are far more intelligent and complex than anything currently known. One name for these hypothetical systems is artilects. With the introduction of artificially intelligent non-deterministic systems, many ethical issues will arise. Many of these issues have never been encountered by humanity.

Over time, debates have tended to focus less and less on "possibility" and more on "desirability", as emphasized in the "Cosmist" (versus "Terran") debates initiated by Hugo de Garis and Kevin Warwick. A Cosmist, according to de Garis, is actually seeking to build more intelligent successors to the human species. The emergence of this debate suggests that desirability questions may also have influenced some of the early thinkers "against".

Designing systems which exceed the intelligence of human beings raises fundamental ethical considerations. Some of these issues are outlined below.

In order to be intelligent does AI need to replicate human thought, and if so, to what extent (eg. can expert systems become AI)? What other avenues to achieving AI exist?
How do we assess the intelligence or sapience of AI?
- Cognition
- Turing test
Can AI be defined in a graded sense (eg. with human-level intelligence graded as 1.0)? What does it mean to have a graduated scale? Is categorisation necessary or important?
AI rights — if AI is comparable in intelligence to humans then they should have comparable rights (as corollary, if AI is more intelligent than humans, would we retain our 'rights'?)
Can AIs be "smarter" than humans in the same way that we are "smarter" than other animals?
Designing and implementing AI 'safeguards'. It is crucial to understand why safeguards should be considered in the first place, however to what extent is it possible to implement safeguards in relation to a superhuman AI? How effective could any such safeguards be?
Some may question the impact upon careers and jobs (eg. there would at least be potential for the problems associated with free trade), however the more crucial issue is the wider impact upon humanity as a whole and human life.
The Singularity

Sub-fields of AI research

GOFAI - 'Good Old Fashioned AI'

Connectionism

Neural network

Artificial Life and Evolution

Modern Bayesian methods and learning

Friendly AI

Friendly AI

Applications

AI in Business

According to Haag, Cummings, etc.(2004) there are four common applications of Artificial Intelligence in the business setting:

Expert Systems
Neural Networks
Genetic Algorithms
Intelligent Agents

Expert Systems apply reasoning capabilities to reach a conclusion. An expert system can process large amounts of known information and provide conclusions based on them.

Neural Networks are AI that are capable of finding and differentiating between patterns. Police Departments use neural networks to identify corruption.

Genetic Algorithms are designed to apply the survival of the fittest process to generate increasingly better solutions to the problem. Investment brokers use Genetic Algorithms to create the best possible combination of investment opportunities for their clients.

An Intelligence Agent is software that assits you, or acts on your behalf, in performing repetitive computer-related tasks. Examples of its uses are data mining programs and monitoring and surveillance agents.

Logic programming was sometimes considered a field of artificial intelligence, but this is no longer the case.

Famous figures

Machines displaying some degree of intelligence

There are many examples of programs displaying some degree of intelligence. Some of these are:

The Start Project (http://www.ai.mit.edu/projects/infolab/) - a web-based system which answers questions in English.
Brainboost (http://www.brainboost.com) - another question-answering system
Cyc, a knowledge base with vast collection of facts about the real world and logical reasoning ability.
Jabberwacky, a learning chatterbot
ALICE, a chatterbot
Alan (http://www.a-i.com/alan1), another chatterbot
ELIZA, a program which pretends to be a psychotherapist, developed in 1966
PAM (Plan Applier Mechanism) - a story understanding system developed by John Wilensky in 1978.
SAM (Script applier mechanism) - a story understanding system, developed in 1975.
SHRDLU - an early natural language understanding computer program developed in 1968-1970.
Creatures, a computer game with breeding, evolving creatures coded from the genetic level upwards using a sophisticated biochemistry and neural network brains.
BBC news story (http://news.bbc.co.uk/1/hi/wales/3521852.stm) on the creator of Creatures latest creation. Steve Grand's Lucy.
AARON (http://www.kurzweilcyberart.com/KCATaaron/STAFsample) - artificial intelligence, which creates its own original paintings, developed by Raymond Kurzweil.
Eurisko - a language for solving problems which consists of heuristics, including heuristics for how to use and change its heuristics. Developed in 1978 by Douglas Lenat.
X-Ray Vision for Surgeons (http://www.ai.mit.edu/projects/medical-vision/) - a group in MIT which researches medical vision.
Neural networks-based programs for backgammon and go (http://www.jellyfish-ai.com).

AI researchers

There are many thousands of AI researchers (http://en.wikipedia.org/wiki/Category:Artificial_intelligence_researchers) around the world at hundreds of research institutions and companies. Among the many who have made significant contributions are:

To some computer scientists, the phrase artificial intelligence has acquired somewhat of a bad name due to the large discrepancy between what has been achieved so far in the field and some more usual notions of intelligence. This problem has been aggravated by various popular science writers and media personalities such as Kevin Warwick whose work has raised the expectations of AI research far beyond its current capabilities. For this reason, some researchers working on topics related to artificial intelligence say they work in cognitive science, informatics, statistical inference or information engineering. However, progress has in fact been made, and AI is today routinely employed in thousands of industrial systems around the world. See Raj Reddy's AAAI paper for a huge review of real-world AI systems in deployment today.

Further reading

Non-fiction

See also Important publications in artificial intelligence.

Artificial Intelligence: A Modern Approach by Stuart J. Russell and Peter Norvig ISBN 0130803022
Gödel, Escher, Bach : An Eternal Golden Braid by Douglas R. Hofstadter
Shadows of the Mind and The Emperor's New Mind by Roger Penrose
Consciousness Explained by Dennett.
The Age of Spiritual Machines by Ray Kurzweil
Understanding Understanding: Essays on Cybernetics and Cognition by Heinz von Foerster
In the Image of the Brain: Breaking the Barrier Between Human Mind and Intelligent Machines by Jim Jubak
Today's Computers, Intelligent Machines and Our Future by Hans Moravec, Stanford University
The Society of Mind by Marvin Minsky, ISBN 0671657135 March 15, 1998
Perceptrons: An Introduction to Computational Geometry by Marvin Minsky and Seymour Papert ISBN 0262631113 December 28, 1987

Fiction

The following is a list of influential works See also longer lists at:-

HAL 9000 in 2001 A Space Odyssey
HARLIE in When H.A.R.L.I.E. was One by David Gerrold
A.I.: Artificial Intelligence
Artificial intelligence - mainly its philosophical implications and its impact on Humanity -- is a major theme in David Lodge's campus novel Thinks ... (2001).
The Moon is a Harsh Mistress by Robert A. Heinlein
William Gibson's Neuromancer
Isaac Asimov's I, Robot series, introducing the famous Three Laws of Robotics, is often considered to be the most accurate fictional depiction of AI

Sources

John McCarthy: Proposal for the Dartmouth Summer Research Project On Artificial Intelligence. [2] (http://www-formal.stanford.edu/jmc/history/dartmouth/dartmouth.html)

See also

Philosophy

Logic

Semantics

Science

Applications

Uncategorised

Collective intelligence - the idea that a relatively large number of people co-operating in one process can lead to reliable action.
Quantum mind - the idea that large-scale quantum coherence is necessary to understand the brain.
the Singularity - a time at which technological progress accelerates beyond the ability of current-day human beings to understand it, or the point in time of the emergence of smarter-than-human intelligence.
Mindpixel - A project to collect simple true / false assertions and collaboratively validate them with the aim of using them as a body of human common sense knowledge that can be utilised by a machine.
Game programming AI
artificial consciousness
truth maintenance systems - by Gerald Jay Sussman and Richard Stallman
K-line (artificial intelligence)

External links

General

Programming:AI (http://wikibooks.org/wiki/Programming:AI) @ Wikibooks.org
University of Berkeley AI Resources (http://www.cs.berkeley.edu/~russell/ai.html) linking to about 869 other WWW pages about AI
Loebner Prize website (http://www.loebner.net/Prizef/loebner-prize.html)
Jabberwacky - a learning AI chatterbot (http://www.jabberwacky.com)
AIWiki (http://purl.net/net/AIWiki) - a wiki devoted to AI.
AIAWiki (http://ai.squeakydolphin.com/) - AI algorithms and research.
AI web category on Open Directory (http://www.dmoz.org/Computers/Artificial_Intelligence/)
Mindpixel (http://www.mindpixel.com/) "The Planet's Largest Artificial Intelligence Effort"
OpenMind CommonSense (http://commonsense.media.mit.edu/cgi-bin/search.cgi/) "Teaching computers the stuff we all know"
Artificially Intelligent Ouija Board (http://www.bitesizeinc.net/index.php/ouija.html) - creative example of human-like AI
Artificial Life (http://www.ifi.unizh.ch/ailab/) - AI Lab, Zurich
Heuristics and AI in finance and investment (http://www.geocities.com/francorbusetti/)
SourceForge Open Source AI projects (http://sourceforge.net/softwaremap/trove_list.php?form_cat=133) - 1139 projects
Ethical and Social Implications of AI en Computerization (http://www.aaai.org/AITopics/html/ethics.html)
AI algorithm implementations and demonstrations (http://www.cs.unm.edu/~luger/ai-final/software.html)
Marvin Minsky's Homepage (http://web.media.mit.edu/~minsky/)
MIT's AI Lab (http://www.ai.mit.edu/)
MIT (http://www.mit.edu)
Informatics Department at University of Sussex (http://www.sussex.ac.uk/informatics/)
Why Programming is a Good Medium for Expressing Poorly Understood and Sloppily Formulated Ideas (http://metainformaciones.blogspot.com/2005/02/minksy-y-la-programacin.html)
aiKnow: Cognitive Artificial Intelligence (http://www.vanschaik.tk/aiknow)

AI related organizations

American Association for Artificial Intelligence (http://www.aaai.org/)
European Coordinating Committee for Artificial Intelligence (http://www.eccai.org/)
The Association for Computational Linguistics (http://www1.cs.columbia.edu/~acl/)
Artificial Intelligence Student Union (http://www.dotmotive.com/~aisu/)
German Research Center for Artificial Intelligence, DFKI GmbH (http://www.dfki.de/)
Association for Uncertainty in Artificial Intelligence (http://www.auai.org/)
Singularity Institute for Artificial Intelligence (http://www.singinst.org)
The Society for the Study of Artificial Intelligence and the Simulation of Behaviour (United Kingdom) (http://www.aisb.org.uk/)

Retrieved from "http://en.wikipedia.org/wiki/Artificial_intelligence"

Categories: Artificial intelligence | Science | Philosophy of mind | Robotics

Views

Personal tools

Create account / log in

Navigation

Search

Toolbox

In other languages