Artificial intelligence: Difference between revisions

Browse history interactively ← Previous edit Next edit →Content deleted Content addedVisual WikitextInline

Revision as of 17:47, 30 November 2007 view sourceCharlesGillingham (talk \| contribs)Extended confirmed users21,415 editsm →AI research: This sentence appeared in two places.← Previous edit		Revision as of 05:20, 1 December 2007 view source Orangemike (talk \| contribs)Autopatrolled, Administrators126,212 editsm →Other sources: remove apparent vanity insertionNext edit →
(5 intermediate revisions by one other user not shown)
Line 288:		Line 288:
	{{Harvnb\|Russell\|Norvig\|2003\|pp=820-824}}		{{Harvnb\|Russell\|Norvig\|2003\|pp=820-824}}
	</ref>		</ref>
			Many researchers hope that a sufficiently powerful natural language processing system would be able to acquire knowledge on it's own, by reading the existing text available over the internet. One young inventor named Daniel Nase put together a program in November 2001 using Boolean that completely replicated the English language. The theory behind this program was "if man reasons primarily through language, then a computer with a complete understand of language could reason and learn just like any human being". This program used a variety of Boolean Logic tests and reduced the English language into a simple Boolean Logic format and used Goto instructions for processing information. One unique aspect of the program was its ability to reason and discard inaccurate information by performing logic tests and emailing a carefully selected group of teachers. Daniel refered to the program as "a developing child, the more you teach it, the more useful it becomes". The prototype of this program was lost due to a breakin at his ocompany office in Auburn, WA. Daniel says, "the program was a thousand lines of code and it took him 3 months to complete working 16 hours a day, and I could write it again". Daniel believed that the mature application would require a mainframe or super computer to run, because of the enormous amount of processing and storage space required. He said "it was literally designed to learn anything and operate any program, but you had to tell it to take action and show it how to do it the first time". "It isn't the Hollywood version of A.I." Daniel envisioned that "this new type of program would replace future operating systems and possibly interact with end users via their cell phones". He also said it would provide consumers with a powerful software program that can quickly learn, adapt and combat malicious software, and the mature program could boost the productivity of software engineers by hundreds or even thousands of times depending on the amount of resources available".
	Many researchers hope that a sufficiently powerful natural language processing system would be able to acquire knowledge on it's own, by reading the existing text available over the internet.

	Some straightforward applications of natural language processing include ] (or ]) and ].<ref>		Some straightforward applications of natural language processing include ] (or ]) and ].<ref>

Revision as of 05:20, 1 December 2007

"AI" redirects here. For other uses of "AI" and "Artificial intelligence", see AI (disambiguation).

File:P11 kasparov breakout.jpg

Garry Kasparov playing against Deep Blue, the first machine to win a chess match against a reigning world champion.

The modern definition of artificial intelligence (or AI) is "the study and design of intelligent agents" where an intelligent agent is a system that perceives its environment and takes actions which maximizes its chances of success. John McCarthy, who coined the term in 1956, defines it as "the science and engineering of making intelligent machines." Other names for the field have been proposed, such as computational intelligence, synthetic intelligence or computational rationality.

The term artificial intelligence is also used to describe a property of machines or programs: the intelligence that the system demonstrates. Among the traits that researchers hope machines will exhibit are reasoning, knowledge, planning, learning, communication, perception and the ability to move and manipulate objects. General intelligence (or "strong AI") has not yet been achieved and is a long-term goal of AI research.

AI research uses tools and insights from many fields, including computer science, psychology, philosophy, neuroscience, cognitive science, linguistics, operations research, economics, control theory, probability, optimization and logic. AI research also overlaps with tasks such as robotics, control systems, scheduling, data mining, logistics, speech recognition, facial recognition and many others.

Perspectives on AI

The rise and fall of AI in public perception

Main articles: History of artificial intelligence and Timeline of artificial intelligence See also: AI Winter

The notion of artificial intelligence dates back to classical antiquity, however it was not until the advent of the modern programmable digital computer that algorithms could be developed and tested. The first coordinated research effort took place at a conference on the campus of Dartmouth College in the summer of 1956. Those who attended would become the leaders of AI research for many decades, especially John McCarthy, Marvin Minsky, Allen Newell and Herbert Simon, who founded AI laboratories at MIT, CMU and Stanford. They and their students wrote programs that were, to most people, simply astonishing: computers were solving word problems in algebra, proving logical theorems and speaking English. By the middle 60s their research was heavily funded by DARPA, and they were optimistic about the future of the new field:

1965, H. A. Simon: "machines will be capable, within twenty years, of doing any work a man can do"
1967, Marvin Minsky: "Within a generation ... the problem of creating 'artificial intelligence' will substantially be solved."

These predictions, and many like them, would not come true. They had failed to recognize the difficulty of some of the problems they faced. In 1974, in response to the criticism of England's Sir James Lighthill and ongoing pressure from congress to fund more productive projects, DARPA cut off all undirected, exploratory research in AI. This was the first AI Winter.

In the early 80s, the field was revived by the commercial success of expert systems and by 1985 the market for AI had reached more than a billion dollars. Minsky and others warned the community that enthusiasm for AI had spiraled out of control and that disappointment was sure to follow. Minsky was right. Beginning with the collapse of the Lisp Machine market in 1987, AI once again fell into disrepute, and a second, more lasting AI Winter began.

In the 90s AI achieved its greatest successes, albeit somewhat behind the scenes. Artificial intelligence was adopted throughout the technology industry, providing the heavy lifting for logistics, data mining, medical diagnosis and many other areas. The success was due to several factors: the incredible power of computers today (see Moore's law), a greater emphasis on solving specific subproblems, the creation of new ties between AI and other fields working on similar problems, and above all a new commitment by researchers to solid mathematical methods and rigorous scientific standards.

The philosophy of AI

Main articles: Philosophy of artificial intelligence and Ethics of artificial intelligence

The strong AI vs. weak AI debate ("can a man-made artifact be conscious?") is still a hot topic amongst AI philosophers. This involves philosophy of mind and the mind-body problem. Most notably Roger Penrose in his book The Emperor's New Mind and John Searle with his "Chinese room" thought experiment argue that true consciousness cannot be achieved by formal logic systems, while Douglas Hofstadter in Gödel, Escher, Bach and Daniel Dennett in Consciousness Explained argue in favour of functionalism. In many strong AI supporters' opinions, artificial consciousness is considered the holy grail of artificial intelligence. Edsger Dijkstra famously opined that the debate had little importance: "The question of whether a computer can think is no more interesting than the question of whether a submarine can swim."

Epistemology, the study of knowledge, also makes contact with AI, as engineers find themselves debating similar questions to philosophers about how best to represent and use knowledge and information (e.g., semantic networks).

AI in myth and fiction

Main article: Artificial intelligence in fiction

Beings created by man have existed in mythology long before their currently imagined embodiment in electronics (and to a lesser extent biochemistry). Some notable examples include: Golems, and Frankenstein. These, and our modern science fiction stories, enables us to imagine that the fundamental problems of perception, knowledge representation, common sense reasoning, and learning have been solved and let's us consider the technology's impact on society. With Artificial Intelligence's theorized potential equal to or greater than our own, the impact can range from service (R2D2), cooperation (Lt. Commander Data), and/or human enhancement (Ghost in the Shell) to our domination (With Folded Hands) or extermination (Terminator (series), The Matrix (series), Battlestar Galactica (re-imagining)). Given the negative consequences, ranging from fear of losing one's job to an AI, the clouding of our self image, to the extreme of the AI Apocalypse, it is not surprising the Frankenstein complex would be a common reaction. Subconsciously we demonstrate this same fear in the Uncanny Valley hypothesis. See AI and Society in fiction for more ...

With the capabilities of a human, a sentient AI can play any of the roles normally ascribed to humans in literature, such as protagonist (Bicentennial Man (film)), antagonist (Terminator, HAL 9000), faithful companion (R2D2), cometic relief (C3PO). See Sentient AI in fiction ...

While most portrayals of AI in science fiction deal with sentient AIs, many imagined futures incorporate AI subsystems in their vision: such as self-navigating cars and speech recognition systems. See non-sentient AI in fiction for more ...

The inevitability of the integration of AI into human society is also argued by some science/futurist writers such as Kevin Warwick and Hans Moravec and the manga Ghost in the Shell

The future of AI

Main article: Strong AI

AI research

Problems of AI

While there is no universally accepted definition of intelligence, AI researchers have studied several traits that are considered essential.

Deduction, reasoning, problem solving

Early AI researchers developed algorithms that imitated the process of conscious, step-by-step reasoning that human beings use when they solve puzzles, play board games, or make logical deductions. These early methods often couldn't be applied to real world situations because the were unable to handle incomplete or imprecise information. However, by the late 80s and 90s, AI research developed highly successful methods for dealing with uncertainty, employing concepts from probability and economics.

For difficult problems, most of these algorithms can require enormous computational resources — most experience a "combinatorial explosion": the amount of memory or computer time required becomes astronomical when the problem goes beyond a certain size. The search for more efficient problem solving algorithms is a high priority for AI research.

It is not clear, however, that conscious human reasoning is any more efficient when faced with a difficult abstract problem. Cognitive scientists have demonstrated that human beings solve most of their problems using unconscious reasoning, rather than the conscious, step-by-step deduction that early AI research was able to model. For many problems, people seem to simply jump to the correct solution: they think "instinctively" and "unconsciously". These instincts seem to involve skills usually applied to other problems, such as motion and manipulation (our so-called "embodied" skills that allow us deal with the physical world) or perception (for example, our skills at pattern matching). It is hoped that sub-symbolic methods, like computational intelligence and situated AI, will be able to model these instinctive skills. The problem of unconscious problem solving, which forms part of our commonsense reasoning, is largely unsolved.

Knowledge representation

Main articles: knowledge representation and commonsense knowledge

Another important measure of intelligence is how much an agent knows. Many of the problems machines are expected to solve will require extensive knowledge about the world. Knowledge representation and knowledge engineering are central to AI research. Among the things that AI needs to represent are: objects, properties, categories and relations between objects; situations, events, states and time; causes and effects; knowledge about knowledge (what we know about what other people know); and many other, less well researched domains. A complete representation of "what exists" is an ontology (borrowing a word from traditional philosophy). Ontological engineering is the science of finding a general representation that can handle all of human knowledge.

Among the most difficult problems in knowledge representation are:

Default reasoning and the qualification problem: Many of the things people know take the form of "working assumptions." For example, if a bird comes up in conversation, people typically picture a animal that is fist sized, sings, and flies. None of these things are true about birds in general. John McCarthy identified this problem in 1969 as the qualification problem: for any commonsense rule that AI researchers care to represent, there tend to be a huge number of exceptions. Almost nothing is simply true or false in the way that abstract logic requires. AI research has explored a number of solutions to this problem.
Unconscious knowledge: Much of what people know isn't represented as "facts" or "statements" that they could actually say out loud. They take the form of intuitions or tendencies and are represented in the brain unconsciously and sub-symbolically. This unconscious knowledge informs, supports and provides a context for our conscious knowledge. As with the related problem of unconscious reasoning, it is hoped that situated AI or computational intelligence will provide ways to represent this kind of knowledge.
The breadth of common sense knowledge: The number of atomic facts that the average person knows is astronomical. Research projects that attempt to build a complete knowledge base of commonsense knowledge, such as Cyc, require enormous amounts of tedious step-by-step ontological engineering — they must be built, by hand, one complicated concept at a time.

Planning

Main article: automated planning and scheduling

Intelligent agents must be able set goals and achieve them. They need a way to visualize the future: they must have a representation of the state of the world and be able to make predictions about how their actions will change it. There are several types of planning problems:

Classical planning problems assume that the agent is the only thing acting on the world, and that the agent can be certain what the consequences of it's actions may be. Partial order planning problems take into account the fact that sometimes it's not important which sub-goal the agent achieves first.
If the environment is changing, or if the agent can't be sure of the results of its actions, it must periodically check if the world matches its predictions (conditional planning and execution monitoring) and it must change its plan as this becomes necessary (replanning and continuous planning).
Some planning problems take into account the utility or "usefulness" of a given outcome. These problems can be analyzed using tools drawn from economics, such as decision theory or decision analysis and information value theory.
Multi-agent planning problems try to determine the best plan for a community of agents, using cooperation and competition to achieve a given goal. These problems are related emerging fields like evolutionary algorithms and swarm intelligence.

Learning

Main article: machine learning

Natural language processing

Main article: natural language processing

Natural language processing gives machines the ability to be read and understand the languages human beings speak. The problem of natural language processing involves such subproblems as syntax and parsing, semantics and disambiguation, and discourse understanding. (e.g., identifying the speech act, using coherence relations in the text, and deciphering the speaker's intentions or pragmatics.) Many researchers hope that a sufficiently powerful natural language processing system would be able to acquire knowledge on it's own, by reading the existing text available over the internet. One young inventor named Daniel Nase put together a program in November 2001 using Boolean that completely replicated the English language. The theory behind this program was "if man reasons primarily through language, then a computer with a complete understand of language could reason and learn just like any human being". This program used a variety of Boolean Logic tests and reduced the English language into a simple Boolean Logic format and used Goto instructions for processing information. One unique aspect of the program was its ability to reason and discard inaccurate information by performing logic tests and emailing a carefully selected group of teachers. Daniel refered to the program as "a developing child, the more you teach it, the more useful it becomes". The prototype of this program was lost due to a breakin at his ocompany office in Auburn, WA. Daniel says, "the program was a thousand lines of code and it took him 3 months to complete working 16 hours a day, and I could write it again". Daniel believed that the mature application would require a mainframe or super computer to run, because of the enormous amount of processing and storage space required. He said "it was literally designed to learn anything and operate any program, but you had to tell it to take action and show it how to do it the first time". "It isn't the Hollywood version of A.I." Daniel envisioned that "this new type of program would replace future operating systems and possibly interact with end users via their cell phones". He also said it would provide consumers with a powerful software program that can quickly learn, adapt and combat malicious software, and the mature program could boost the productivity of software engineers by hundreds or even thousands of times depending on the amount of resources available".

Some straightforward applications of natural language processing include information retrieval (or text mining) and machine translation.

Perception

Main articles: machine perception, computer vision, and speech recognition

Motion and manipulation

Main article: robotics

Social intelligence

Main article: affective computing

Emotion and social skills play two roles for an intelligent agent:

It must be able predict the actions of others, by understanding their motives and emotional states. (This involves elements of game theory, decision theory, as well as the ability model human emotions and the perceptual skills to detect emotions.)
For good human-computer interaction, an intelligent machine also to needs to display emotions — at the very least it must appear polite and sensitive to the humans it interacts with. At best, it should appear to have normal emotions itself.

General intelligence

Main articles: strong AI and AI-complete

Most researchers hope that their work will eventually be incorporated into a machine with general intelligence (known as strong AI), combining all the skills above and exceeding human abilities at most or all of them. A few believe that anthropomorphic features like artificial consciousness or an artificial brain may be required for such a project.

Many of the problems above are considered AI-complete: to solve one problem, you must solve them all. For example, even a straight-forward, limited and specific task like machine translation is AI complete. To translate accurately, a machine must be able to understand the text. It must be able to follow the author's argument, so it must have some ability to reason. It must have extensive world knowledge so that it knows what is being discussed — it must at least be familiar with all the same commonsense facts that the average human translator knows. Some of this knowledge is in the form of facts that can be explicitly represented, but some knowledge is unconscious and closely tied to the human body: for example, the machine may need to understand how an ocean makes one feel to accurately translate a specific metaphor in the text. It must also model the authors' goals, intentions, and emotional states to accurately reproduce them in a new language. In short, the machine is required to have wide variety of human intellectual skills, including reasoning, commonsense knowledge and the intuitions that underly motion and manipulation, perception, and social intelligence. Machine translation, therefore, is believed to be AI-complete: it may require strong AI to be done as well as humans can do it.

Approaches to AI

Artificial intelligence is a young science and is still a fragmented collection of subfields. At present, there is no established unifying theory that links the subfields into a coherent whole.

Cybernetics and brain simulation

In the 40s and 50s, a number of researchers explored the connection between neurology, information theory, and cybernetics. Some of them built machines that used electronic networks to exhibit rudimentary intelligence, such as W. Grey Walter's turtles and the Johns Hopkins Beast. Many of these researchers gathered for meetings of the Teleological Society at Princeton and the Ratio Club in England.

Traditional symbolic AI

When access to digital computers became possible in the middle 1950s, AI research began explore that possibility that human intelligence could be reduced to symbol manipulation. The research was centered in three institutions: CMU, Stanford and MIT, and each one developed it's own style of research. John Haugeland named these approaches to AI "good old fashioned AI" or "GOFAI".

Cognitive simulation: Economist Herbert Simon and Alan Newell studied human problem solving skills and attempted to formalize them, and their work laid the foundations of the field of artificial intelligence, as well as cognitive science, operations research and management science. Their research team performed psychological experiments to demonstrate the similarities between human problem solving and the programs (such as their "General Problem Solver") they where developing. This tradition, centered at Carnegie Mellon University, would eventually culminate in the development of the Soar architecture in the middle 80s.

Logical AI: Unlike Newell and Simon, John McCarthy felt that machines did not need to simulate human thought, but should instead try find the essence of abstract reasoning and problem solving, regardless of whether people used the same algorithms. His laboratory at Stanford (SAIL) focussed on using formal logic to solve wide variety of problems, including knowledge representation, planning and learning. Work in logic led to the development of the programming language Prolog and the science of logic programming.

"Scruffy" symbolic AI: In contrast to the formal methods pursued at CMU, Stanford and Edinburgh, the researchers at MIT (such as Marvin Minsky and Seymour Papert) found that solving difficult problems in vision and natural language processing required ad-hoc solutions -- they argued that there was no silver bullet, no simple and general principle (like logic) that would capture all the aspects of intelligent behavior. An important realization was that AI required large amounts of commonsense knowledge, and that this had to be engineered one complicated concept at time. This tradition, which Roger Schank named "scruffy AI" still forms the basis of research into commonsense knowledge, such as Doug Lenat's Cyc.

Knowledge based AI: When computers with large memories became available around 1970, researchers from all three traditions began to build knowledge into AI applications. This "knowledge revolution" led to the development and deployment of expert systems, the first truly successful form of AI software.

Sub-symbolic AI

During the 1960s, symbolic approaches had achieved great success at simulating high-level thinking in small demonstration programs. Approaches based on cybernetics or neural networks were abandoned or pushed into the background. By the 1980s, however, progress in symbolic AI seemed to stall and many believed that symbolic systems would never be able to imitate all the processes of human cognition, especially perception, robotics, learning and pattern recognition. A number of researchers began to look into "sub-symbolic" approaches to specific AI problems.

Bottom-up, situated, behavior based or nouvelle AI: Researchers from the related field of robotics, such as Rodney Brooks, rejected symbolic AI and focussed on the basic engineering problems that would allow robots to move and survive. Their work revived the non-symbolic viewpoint of the early cybernetics researchers of the 50s and reintroduced the use of control theory in AI. These "bottom-up" approaches are known as behavior-based AI, situated AI or Nouvelle AI.

Computational Intelligence: Interest in neural networks and "connectionism" was revived by David Rumelhart and others in the middle 1980s. These and other sub-symbolic approaches, such as fuzzy systems and evolutionary computation, are now studied collectively by the emerging discipline of computational intelligence.

The new neats: In the 1990s, AI researchers developed sophisticated mathematical tools to solve specific subproblems. These tools are truly scientific, in the sense that their results are both measurable and verifiable, and they have been responsible for many of AI's recent successes. The shared mathematical language has also permitted a high level of collaboration with more established fields (like mathematics, economics or operations research). Russell & Norvig (2003) describe this movement as nothing less than a "revolution" and "the victory of the neats."

Intelligent agent paradigm

The "intelligent agent" paradigm became widely accepted during the 1990s. Although earlier researchers had proposed modular "divide and conquer" approaches to AI, the intelligent agent did not reach its modern form until Judea Pearl, Alan Newell and others brought concepts from decision theory and economics into the study of AI. When the economist's definition of a rational agent was married to computer science's definition of an object or module, the intelligent agent paradigm was complete.

An intelligent agent is a system that perceives its environment and takes actions which maximizes its chances of success. The simplest intelligent agents are programs that solve specific problems. The most complicated intelligent agents would be rational, thinking human beings.

The paradigm gives researchers license to study isolated problems and find solutions that are both verifiable and useful, without agreeing on single approach. An agent that solves a specific problem can use any approach that works — some agents are symbolic and logical, some are sub-symbolic neural networks and some can be based on new approaches (without forcing researchers reject old approaches that have been proven to work). The paradigm provides a common language to describe problems and share their solutions with each other, and with other fields that also used concepts of abstract agents, like decision theory.

Integrating the approaches

An agent architecture or cognitive architecture allows researchers to build more versatile and intelligent systems out of interacting intelligent agents in a multi-agent system. A system with both symbolic and sub-symbolic components is a hybrid intelligent system, and the study of such systems is artificial intelligence systems integration.

Tools of AI research

In the course of 50 years of research, AI has developed a large number of tools to solve the most difficult problems in computer science. A few of the most general of these methods are discussed below.

Search

Main article: search algorithm

Many problems in AI can be solved in theory by intelligently searching through many possible solutions:

Reasoning can be reduced to a form a search. For example, in game playing, the agent can search through a tree of possible moves and counter moves to find a strategy that improves its position. (Tools for two person games include minimax and alpha-beta pruning.) Logical proof can be viewed as searching for a path the leads from premises to conclusions, where each step is the application of an inference rule. Many other reasoning problems, such as constraint satisfaction and dynamic programming are solved using a form of search.
Planning algorithms search through trees of goals and subgoals, attempting to find a path a target goal. These sets of goals and subgoals can be represented with graphs (as in the graphplan algorithm), or in a hierarchical task network.

There a several types of search algorithms:

"Naive" search algorithms, such as breadth first search, depth first search and general state space search.
Heuristic or "informed" search. The naive algorithms quickly run into problems: the search space quickly can expand to astronomical size in a "combinatorial explosion."AI researchers realized from the beginning that they would need to trim the search space using heuristics or "rules of thumb". The use of heuristics led to the development of intelligent searches such as greedy best first and A*.
Local searches, such as hill climbing, simulated annealing and beam search, use techniques borrowed from optimization theory.
Genetic algorithms are a form of optimization search that imitates the process of natural selection, searching for an artificial phenotype (i.e. any sort of pattern) which passes a fitness measure by producing many copies of the most successful versions (imitating inheritance) and modifying them slightly (imitating the process of mutation).

Logic

Main articles: logic programming and production system

Logic was introduced into AI research by John McCarthy in his 1958 Advice Taker proposal. The most important technical development was J. Alan Robinson's discovery of the resolution and unification algorithm for logical deduction in 1963. This procedure is simple, complete and entirely algorithmic, and can easily be performed by digital computers. However, a naive implementation of the algorithm quickly leads to a combinatorial explosion or an infinite loop, so sophisticated search methods are used to implement the inference engine that is at the core of a logical agent or logic programming system.

Logic is used for knowledge representation and problem solving, but it can applied to other problems as well. For example, the satplan algorithm uses logic for planning, and inductive logic programming is a method for learning.

There are several different forms of logic used in AI research.

Propositional logic or sentential logic is the logic of statements which can be true or false.
First order logic also allows the use of quantifiers and predicates, and can express facts about objects, their properties, and their relations with each other.
A production system is a version of first order logic in which the only form of statement allowed is a rule of the form "if p then q" (i.e., a Horn clause) or an atomic fact (such as "IsA"). This reduces the process of inference to forward and backward chaining.
Fuzzy logic, a version of first order logic which allows the truth of statement to represented as a value between 0 and 1, rather than simply True (1) or False (0). Fuzzy systems can be used for uncertain reasoning and have been widely used in modern industrial and consumer product control systems.
Default logics, non-monotonic logics and circumscription are forms of logic designed to help with default reasoning and the qualification problem.
Several extensions of logic have been designed to handle specific domains of knowledge, such as description logics, situation calculus, event calculus and fluent calculus (for representing events and time), causal calculus, belief calculus and modal logics.

Stochastic methods

Main articles: Bayesian network, hidden Markov model, and Kalman filter

Starting in the late 80s and early 90s, Judea Pearl and others championed the use of stochastic or probabilistic methods in artificial intelligence. Researchers have used principles from probability theory to devise a number of powerful tools.

Bayesian networks have been applied to a large number of problems, such as: uncertain reasoning (using the Bayesian inference algorithm), learning (using the expectation-maximization algorithm), and planning (using decision networks).

Probabilistic methods have been particularly successful at dealing with processes that occur over time. Several successful algorithms have been developed for filtering, prediction, smoothing and finding explanations for streams of data, such as hidden Markov models, Kalman filters and dynamic Bayesian networks. These tools are used for the problems of perception (such as pattern matching) and learning.

Economic models

Main articles: utility theory, decision theory, and game theory

AI has been able to use tools drawn from economics, such as decision theory and decision analysis, Bayesian decision networks, information value theory, Markov decision processes, dynamic decision networks, game theory and mechanism design These tools have been especially important for planning problems.

Classifiers and statistical learning methods

Main articles: classifier (mathematics) and machine learning

The simplest AI applications can be divided into two types: classifiers ("if shiny then diamond") and controllers ("if shiny then pick up"). Controllers do however also classify conditions before inferring actions, and therefore classification forms a central part of many AI systems.

Classifiers are functions that use pattern matching to determine a closest match. They can be tuned according to examples, making them very attractive for use in AI. These examples are known as observations or patterns. In supervised learning, each pattern belongs to a certain predefined class. A class can be seen as a decision that has to be made. All the observations combined with their class labels are known as a data set.

When a new observation is received, that observation is classified based on previous experience. A classifier can be trained in various ways; there are mainly statistical and machine learning approaches.

A wide range of classifiers are available, each with its strengths and weaknesses. Classifier performance depends greatly on the characteristics of the data to be classified. There is no single classifier that works best on all given problems; this is also referred to as the "no free lunch" theorem. Various empirical tests have been performed to compare classifier performance and to find the characteristics of data that determine classifier performance. Determining a suitable classifier for a given problem is however still more an art than science.

The most widely used classifiers are the neural network, kernel methods such as the support vector machine, k-nearest neighbor algorithm, Gaussian mixture model, naive Bayes classifier, and decision tree. The performance of these classifiers have been compared over a wide range of classification tasks in order to find data characteristics that determine classifier performance.

Neural networks

Main articles: neural networks and connectionism

Techniques and technologies in AI which have been directly derived from neuroscience include neural networks and connectionism. The study of neural networks began with cybernetics researchers, working in the decade before the field AI research was founded. In the 1960s Frank Rosenblatt developed an important early version, the perceptron. Paul Werbos discovered the backpropagation algorithm in 1984, which led to a renaissance in neural network research and connectionism in general in the middle 1980s. The Hopfield net, a form of attractor network, was first described by John Hopfield in 1982.

Neural networks are applied to the problem of learning, using such techniques as Hebbian learning and the relatively new field of Hierarchical Temporal Memory which simulates the architecture of the neocortex.

Social and emergent models

Main article: evolutionary computation

Several algorithms for learning use tools from evolutionary computation, such as genetic algorithms and swarm intelligence.

Control theory

Main article: intelligent control

Control theory, the grandchild of cybernetics, has many important applications, especially in robotics.

Specialized languages

Main articles: IPL, Lisp (programming language), Prolog, STRIPS, and Planner

AI researchers have developed specialized languages for AI research. The first programming language developed for AI was IPL, developed by Alan Newell, Herbert Simon and J. C. Shaw, and the two most important languages are Lisp (developed by John McCarthy at MIT in 1958) and Prolog, a language based on logic programming (invented by French researchers Alain Colmerauer and Phillipe Roussel, in collaboration with Robert Kowalski of the University of Edinburgh). These two languages are used in the vast majority of AI applications. Two other important historical languages are the planning languages STRIPS (developed at Stanford) and Planner (developed at MIT), both in the 1960s.

Research challenges

A legged league game from RoboCup 2004 in Lisbon, Portugal.

The 800 million-Euro EUREKA Prometheus Project on driverless cars (1987-1995) showed that fast autonomous vehicles, notably those of Ernst Dickmanns and his team, can drive long distances (over 100 miles) in traffic, automatically recognizing and tracking other cars through computer vision, passing slower cars in the left lane. But the challenge of safe door-to-door autonomous driving in arbitrary environments will require additional research.

The DARPA Grand Challenge was a race for a $2 million prize where cars had to drive themselves over a hundred miles of challenging desert terrain without any communication with humans, using GPS, computers and a sophisticated array of sensors. In 2005, the winning vehicles completed all 132 miles (212 km) of the course in just under seven hours. This was the first in a series of challenges aimed at a congressional mandate stating that by 2015 one-third of the operational ground combat vehicles of the US Armed Forces should be unmanned. For November 2007, DARPA introduced the DARPA Urban Challenge. The course will involve a sixty-mile urban area course. Darpa has secured the prize money for the challenge as $2 million for first place, $1 million for second and $500,000 for third.

A popular challenge amongst AI research groups is the RoboCup and FIRA annual international robot soccer competitions. Hiroaki Kitano has formulated the International RoboCup Federation challenge: "In 2050 a team of fully autonomous humanoid robot soccer players shall win the soccer game, comply with the official rule of the FIFA, against the winner of the most recent World Cup."

A lesser known challenge to promote AI research is the annual Arimaa challenge match. The challenge offers a $10,000 prize until the year 2020 to develop a program that plays the board game Arimaa and defeats a group of selected human opponents.

In the post-dot-com boom era, some search engine websites use a simple form of AI to provide answers to questions entered by the visitor. Questions such as What is the tallest building? can be entered into the search engine's input form, and a list of answers will be returned. AskWiki is an example this sort of search engine.

Applications of artificial intelligence

Business

Banks use artificial intelligence systems to organize operations, invest in stocks, and manage properties. In August 2001, robots beat humans in a simulated financial trading competition (BBC News, 2001). A medical clinic can use artificial intelligence systems to organize bed schedules, make a staff rotation, and provide medical information. Many practical applications are dependent on artificial neural networks, networks that pattern their organization in mimicry of a brain's neurons, which have been found to excel in pattern recognition. Financial institutions have long used such systems to detect charges or claims outside of the norm, flagging these for human investigation. Neural networks are also being widely deployed in homeland security, speech and text recognition, medical diagnosis (such as in Concept Processing technology in EMR software), data mining, and e-mail spam filtering.

Robots have become common in many industries. They are often given jobs that are considered dangerous to humans. Robots have proven effective in jobs that are very repetitive which may lead to mistakes or accidents due to a lapse in concentration and other jobs which humans may find degrading. General Motors uses around 16,000 robots for tasks such as painting, welding, and assembly. Japan is the leader in using and producing robots in the world. In 1995, 700,000 robots were in use worldwide; over 500,000 of which were from Japan.

Toys and games

The 1990s saw some of the first attempts to mass-produce domestically aimed types of basic Artificial Intelligence for education, or leisure. This prospered greatly with the Digital Revolution, and helped introduce people, especially children, to a life of dealing with various types of AI, specifically in the form of Tamagotchis and Giga Pets, the Internet (example: basic search engine interfaces are one simple form), and the first widely released robot, Furby. A mere year later an improved type of domestic robot was released in the form of Aibo, a robotic dog with intelligent features and autonomy.

List of applications

Typical problems to which AI methods are applied

Template:MultiCol

| class="col-break " |

Template:EndMultiCol

Other fields in which AI methods are implemented