Artificial intelligence, also
known as machine intelligence, is defined as
intelligence
exhibited by anything manufactured (i.e.
artificial) by
humans or other sentient beings or systems (should such things ever exist
on
Earth or
elsewhere). It is
usually hypothetically applied to general-purpose
computers. The
term is also used to refer to the field of scientific investigation into
the plausibility of and approaches to creating such systems.
Overview
The question of what artificial intelligence is
can be reduced to two parts: "what is the nature of artifice" and "what
is intelligence"? The first question is fairly easy to answer, though it
does point to the question of what it is possible to manufacture (within
the constraints of certain types of system, e.g. classical computational
systems, of available processes of manufacturing and of possible limits
on human intellect, for instance).
The second is much harder, raising questions of
consciousness and
self,
mind (including
the
unconscious mind)
and the question of what components are involved in the only type of
intelligence it is
universally agreed we have available to study: that of human beings.
Intelligent behavior in humans is complex and difficult to study or
understand. Study of animals and artificial systems that are not just
models of what exists already are also considered widely pertinent.
Several distinct types of artificial
intelligence have been elucidated below. Also, the subject divisions,
history, proponents and opponents and applications of research in the
subject are described. Finally, 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
1955 is
"making a machine behave in ways that would be called intelligent if a
human were so behaving." 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 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.
Philosophical criticism and support of strong
AI
Several philosophers, notably
John Searle and
Hubert Dreyfus,
have argued on philosophical grounds against the feasibility of building
human-like 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
What Computers Still Can't Do: A Critique of Artificial Reason, 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.
Other philosophers hold opposing views. Many see
no problem with Weak AI but there is much support for Strong AI too.
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.
Some philosophers hold that if Weak AI is
accepted as possible then so must Strong AI. The Weak AI position, that
intelligence might be apparent but would not be real, is debunked in many
ways, but one accessible example can be found in
Simon Blackburn's
introduction to philosophy, Think. Blackburn points out that
you might appear intelligent but there is no way of telling if that
intelligence is real (whatever that means in this context): We
have to take it on trust or faith.
Supporters of Strong AI claim that the anti-AI
argument boils down in the end to some combination of
-
arrogance as a privileged position is
claimed, a magic spark is introduced (by
God, for
instance)
-
defining
intelligence
as that of which machines are incapable.
An
argument
supporting Strong AI which those who deny its possibility must
necessarily attack:
-
Given that
-
the mind is a
finite state machine
(so the the
Church-Turing thesis
applies to the brain)
-
the mind is software (a
finite state machine)
-
the brain is purely hardware (i.e. only
follows the rules of a classical computer)
-
it exists exclusively within the brain
-
The possibility of Strong AI must be
accepted.
Some (including
Roger Penrose)
attack the applicability of the Church-Turing thesis. Others say the mind
is not completely physical.
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
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:
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".
Some issues that bring up interesting ethical
questions are:
-
Determining the sentience of a system we
create.
-
Can AI be defined in a graded sense?
-
Freedoms and rights for these systems
-
Can AIs be "smarter" than humans in the same
way that we are "smarter" than other animals?
-
Designing systems that are far more
intelligent 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 (e.g.
expert systems)
-
The Singularity
-
Effect on careers and jobs. The problems may
resemble problems seen under
free trade.
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 on English.
-
Cyc, a knowledge
base with vast collection of facts about the real world and logical
reasoning ability.
-
ALICE, a
chatterbot
-
Alan
(http://www.a-i.com/alan1),
another chatterbot
-
ELIZA, a program
which pretends to be a psychotherapist, developed circa 1970.
-
PAM (Plan Applier Mechanism) - a story
understanding system developed by John Wilensky in 1978.
-
SAM (Script applier mechanism) - a story
understanding system, dveloped 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.
-
EURISKO - a
language for solving problems which consists of heuristics, including
heuristics describing 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 progams for backgammon and
go .
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
aggrevated by various irresponsible popular science writers and media
personalities such as
Kevin Warwick
whose work has raised the expectations of AI research way above 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.
Resources
Further Reading
Non-Fiction
Fiction
See also
List of fictional robots and androids
AI related organizations
Sources
Sub-fields of AI research
Logic programming
was sometimes considered a field of artificial intelligence, but this is
no longer the case.
Philosophy
Logic
Applications
Uncategorised
External links
