- This article is about intelligence exhibited by manufactured systems, typically computers. For other uses of the term
AI, see Ai. For the influential Warp Records series, see Artificial Intelligence (series).
Artificial intelligence, also known as machine intelligence, is defined as intelligence exhibited by any manufactured (ie. 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.
Overview
The concept of "artificial intelligence" is something which can be considered in two parts: "what is the nature of artifice"
and "what is intelligence"? The first question is relatively easy to answer, although it also necessarily leads to an examination
of what it is possible to manufacture. For example, the limitations of certain types of systems, such as classical computational
systems, or of available manufacturing processes, or of human intellect, may all place
constraints on what can be manufactured.
The second question raises fundamental ontological issues of consciousness and self and mind (including the unconscious
mind). It also raises questions about the nature of intelligence as displayed by humans, as intelligent behavior in humans is complex and often difficult
to study or understand. Study of animals and artificial systems which are not simply models of what already exists are also
considered highly relevant.
Several distinct types of artificial intelligence are discussed below, along with divisions, history, proponents, opponents,
and applied research in the field. Lastly, references to fictional and non-fictional descriptions of AI are provided.
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 artificial intelligence research 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 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?
- 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
Artificial Life and Evolution
Modern Bayesian methods and learning
Friendly AI
Applications
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 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.
Fiction
The following is a list of influential works See also longer lists at:-
-
Sources
- John McCarthy: Proposal for the Dartmouth Summer Research Project On Artificial Intelligence. [2] (http://www-formal.stanford.edu/jmc/history/dartmouth/dartmouth.html)
Philosophy
Logic
Science
Applications
Uncategorised
External links
General
AI related organizations
|
