Artificial Intelligence

Domain	Definition
Computing	Artificial intelligence (AI) The subfield of computer science concerned with the concepts and methods of symbolic inference by computer and symbolic knowledge representation for use in making inferences. AI can be seen as an attempt to model aspects of human thought on computers. It is also sometimes defined as trying to solve by computer any problem that a human can solve faster. Examples of AI problems are computer vision (building a system that can understand images as well as a human) and natural language processing (building a system that can understand and speak a human language as well as a human). These may appear to be modular, but all attempts so far (1993) to solve them have foundered on the amount of context information and "intelligence" they seem to require. The term is often used as a selling point, e.g. to describe programming that drives the behaviour of computer characters in a game. This is often no more intelligent than "Kill any humans you see; keep walking; avoid solid objects; duck if a human with a gun can see you". See also AI-complete, neats vs. scruffies, neural network, genetic programming, fuzzy computing, artificial life. ACM SIGART (http://sigart.acm.org/). U Cal Davis (http://phobos.cs.ucdavis.edu:8001). CMU Artificial Intelligence Repository (http://www.cs.cmu.edu/Web/Groups/AI/html/repository.html). (2002-01-19). Source: The Free On-line Dictionary of Computing.
Health	The study and implementation of techniques and methods for designing computer systems to perform functions normally associated with human intelligence, such as understanding language, learning, reasoning, problem solving, etc. (references)
Source: compiled by the editor from various references; see credits.

Specialty Definition: Artificial intelligence

AI redirects here, for alternate uses, see AI (disambiguation)

Artificial intelligence, commonly abbreviated as AI, also known as machine intelligence, is the practice of developing algorithms that make machines (usually computers) able to make seemingly intelligent decisions, or act as if possessing intelligence of a human scale.

Overview

The accepted definition of artificial intelligence, put forth by John McCarthy in 1955: "making a machine behave in ways that would be called intelligent if a human were so behaving." Since that time several distinct types of artificial intelligence have been elucidated.

Strong artificial intelligence

Strong artificial intelligence deals with the creation of some form of computer-based artificial intelligence that can truly reason and solve problems; a strong form of AI is said to be sentient, or self-aware. In theory, there are two types of strong AI:

Human-like AI, in which the computer program thinks and reasons much like a human mind.
Non-human-like AI, in which the computer program develops a totally non-human sentience, and a non-human way of thinking and reasoning.

Weak artificial intelligence

Weak artificial intelligence deals with the creation of some form of computer-based artificial intelligence that cannot truly reason and solve problems; such a machine would, in some ways, act as if it were intelligent, but it would not possess true intelligence or sentience.

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.

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. 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].

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.

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 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, and Nathan Rochester of IBM.

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 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:

Deep Blue, a chess-playing computer, beat Garry Kasparov in a famous match in 1997.
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.

The vision of artificial intelligence replacing human professional judgment has arisen many times in the history of the field, 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.

Philosophical criticisms of AI

Several philosophers, notably John Searle and Hubert Dreyfus, have argued on philosophical grounds against the feasibility of building human-link consciousness or intelligence in a disembodied machine. Searle is most known for his Chinese room argument, which claims to demonstrate that even a machine that passed the Turing test would not necessarily be conscious in the human sense. Dreyfus, in his book Why Computers Can't Think, has argued that consciousness cannot be captured by rule- or logic-based systems or by systems that are not attached to a physical body, but leaves open the possibility that a robotic system using neural networks or similar mechanisms might achieve artificial intelligence.

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".

Some issues that bring up interesting ethical questions are:

Determining the sentience of a system we create.
- Turing test
- Cognition
- Why do we have a need to categorize these systems at all
Can AI be defined in a graded sense?
Freedoms and rights for these systems.
Designing systems that are far more impressive than any one human
Deciding how much safe-guards to design into these systems
Seeing how much learning capability a system needs to replicate human thought, or how well it could do tasks without it (eg expert system)
The Singularity

Sub-fields of AI research

Combinatorial search
Computer vision
Expert system
Genetic programming
Genetic algorithm
Knowledge representation
Machine learning
Machine planning
Neural network
Natural language processing
Robotics
Artificial life
Distributed Artificial Intelligence
Swarm Intelligence

Computer programs displaying some degree of "intelligence"

ALICE
ELIZA (demonstration of the limits of AI)
HAL
Mind-1.1