Artificial intelligence

artificial intelligence (short AI or AI, of the English Artificial Intelligence) is a subsection of computer science with an interdisciplinary character. A goal of the AI research is the development of machines with intelligent behavior.

In the understanding of the term reflects artificial intelligenceitself often the conception of humans as machine, originating from the clearing-up , again, whose imitation sits down the so-called strong AI to the goal: an intelligence to create, like humans creatively think and problems solve can and itselfby a form of consciousness and/or self-confident its as well as emotions distinguishes. The goals of the strong AI remained unbeeindruckt after decades of the research illusionaryly and of technical progress.

Contrary to the strong AI it concerns to the weak AI, concrete application problemsto master. In particular thereby such uses of interest are, for whose solution after general understanding a form of „intelligence seems to be necessary “. In the long run it concerns to the weak AI thus the simulation of intelligent behavior with means of mathematics andto computer science, you do not go around creation of consciousness or around a deeper understanding from intelligence. While the strong AI failed because of its philosophical question until today, meaning are on the side of the weak AI in the last yearsProgress obtained.

Beside the research results of the basic informatics into the AI results of the psychology and neurology, mathematics and logic, communication science, philosophy and linguistics flowed. The influence of the neurology has itself in the training of the range Neuro computer science shown, biooriented computer science is assigned to which. Additional also the whole branch of the cognitive science is to be called, which relies substantially on the results of artificial intelligence in co-operation with the cognitive psychology.

Table of contents

of subsections of the AI

at least four kinds of intelligence are differentiated:

  1. Visual intelligence
  2. linguistic intelligence
  3. Manipulative intelligence
  4. rational intelligence

of the moreover one, depending upon differentiation degrees, such intelligence types are added like emotional intelligence. According to four kinds of intelligence are to time of four subsections in the development:

  1. The range of the pattern recognition made devices possible, which can recognize pictures and/or forms, for example finger marks during the crime prevention, the human iris during the person identification, workpieces with the machine manufacturing.
  2. One can nowadays by computersan entered text into language convert and turned around language in as text read. The speech synthesis and speech recognition can function as interface between computers and humans.
  3. In the production technology increasingly freely programmable automats are used, the dangerous works to take over and for example weldingand paint work or monotonous „handles “accomplish.
  4. Computers, those in these ranges to be used are called expert systems. Such expert systems are based on a data base, in which specialized knowledge is stored. Based on it the system can, together with which users solve, specialized tasks. Importantlyit is that each conclusion of the program can be justified by this on the basis the before stored facts. They are used to time in the following ranges with success:
    • medical computer diagnosis
    • of error search and elimination of errors programs
    • industrielle large manufacturing, with the military, civilian aviation, traffic.

The basic ideait is to be examined under which conditions computers can reconstruct behaviors of organisms, which are based on intelligence. Research ranges for this are z. B. the robotics, the knowledge processing and speech recognition.

methods of the AI

the methods of the AI leavearrange themselves roughly in two dimensions: Symbolic one vs. Neural AI and simulation method vs. phenomenological method. Following graphics illustrates the connections:

Zur Einordnung von KI-Methoden und ihren Zusammenhängen

The neural AI pursues bottom UP - a beginning and would like the human brain as precisely as possible to copy. The symbolic AI pursuesturned around one top down - beginning and approaches to the intelligence achievements from a conceptual level. The simulation method orients itself as near as possible at the actual cognitive processes of humans. On the other hand it depends the phenomenological beginning only on the result.

Many older methods, which were developed in the AI, are based on heuristic solution procedures. In recent time mathematically founded beginnings from the statistics , mathematical programming, and the theory of approximation play an important role.

The concrete techniques of the AI leaveare divided roughly in groups:

searches

the AI concerns itself frequently with problems, for which for certain solutions one searches. Different search algorithms are used thereby. A prime example for the search is the procedure of the way identification, inmany computer games takes a central role and on search algorithms like e.g. the A*-Algorithmus is based.

planning an important aspect of the AI represents tarpaulins apart from looking for solutions. The procedure of planning is divided thereby into2 phases:

  1. The formulation of goal: On the basis of the momentary world condition a goal is defined. A goal is here a quantity of world conditions with the one certain goal descriptor is fulfilled.
  2. The formulation of problem: After are which goals admit be aimed at should in the formulation of problem specified which actions and world conditions to be regarded are. Here different problem types exist.

Agent systems plan and provide from such problem descriptions action sequences, which can implement them, in order to achieve their goals.

optimization methods

often leadSetting of tasks for the AI to optimization problems. These are solved depending upon structure either with search algorithms from computer science or, increasingly, with means of mathematical programming. Well-known heuristic search methods from the context of the AI are evolutionary algorithms.

Logical reasoning

a question of the AI is the production of knowledge representations, for automatic logical reasoning to be then used can. Human knowledge is formalized thereby - as far as possible -, in order to bring it into a machine-readable form. This goalthe developers of various Ontologien used up themselves.

Already early the AI was occupied with it to design automatic proof systems (deduction systems) mathematicians and computer scientists when proving sentences and in programming (logic programming) would be helpful. Two difficulties placeditself:

  1. If one formulates sentences in powerful, for which users consent specification languages (for example predicate calculus), the developing search problems become very difficult. In practice one made compromises, where the specification language for the user somewhat pedantically, the associated optimization problems for the computerto handle simpler were (pro log, expert systems).
  2. Even powerful specification languages become unmanageable, if one tries to formulate uncertain or incomplete knowledge. For practical problems this can be a very serious restriction. The current research examines therefore systems, thoseuses the rules of the probability calculation, in order to model Unwissen and uncertainty explicitly. Algorithmically these methods differ much from the older procedures (instead of symbols probability distributions are manipulated).

it

concerns to approximation methods in many applications, out of a quantityof data a general rule to derive (machine learning). Mathematically this leads to an approximation problem. In the context of the AI for this artificial ones neural ones of nets were suggested. In practical applications one uses frequently alternative procedures to analyze mathematically more simplyare.

applications

in the past often changed over realizations of artificial intelligence with the time into the other areas of computer science: As soon as a problem was understood well enough the AI new setting of tasks turned. For examplethe building of compilers or computer algebra was originally added to artificial intelligence.

Numerous applications were developed on the basis by techniques, which were once Forschungsgebiete of the AI or are still it. Some examples:

Turing test

overa measure to have, when a machine simulated humans an equivalent intelligence, was suggested by Alan Turing the Turing test designated after it. Humans place by terminals (thus without view and/or. Hearing contact to the participants) another humansand a AI arbitrary questions. The interrogator must decide thereafter, who from the two asked ones humans are. If humans are not to be differentiated from the machine to, then the machine is intelligent according to Turing, or humans are not it.So far no machine existed this Turing test. Since 1990 the Loebner price for the Turing test exists.

history of the AI

based on the work of Alan Turing (among other things the essay Computing machinery and intelligence) formulated all Newell (1927 - 1992) and harsh ore Simon (1916 - 2001) of the Carnegie Mellon University in Pittsburgh the Physical symbol system Hypothesis, after which thinking is data processing, data processing a computing method, thus symbol manipulation, is and it on the brain assuch when thinking does not arrive: Intelligence is mind implemented by any patternable child OF more matt.

This view that intelligence is independent of the support, is divided thesis by the representatives of the strong AI -, as for example Marvin Minsky (*1927) of the MassachusettsInstitutes OF Technology (WITH), one of the pioneers of the AI, for which „the goal of the AI is the overcoming of death “, or from the robot specialist Hans Moravec (* 1948) of the Carnegie Mellon University, that in its book “Mind Children” (children of theSpirit) the scenario of the evolution of the post office-biological life describes: A robot transfers the knowledge stored in the human brain into a computer, so that the biomass of the brain redundantly will and an post office-human age can begin, in that the stored knowledgearbitrarily for a long time remains accessible.

In particular the initial phase of the AI was coined/shaped by a nearly boundless expectation attitude regarding the ability of computers to solve „tasks for whose solution intelligence is necessary, if they are accomplished by humans “(Minsky). Simonif 1957 prognosticated among other things the fact that within the next 10 years a computer chess world champions become and discover an important mathematical sentence and would prove, prognoses, which did not apply and which Simon 1990, however without date, repeated. It nevertheless succeeded 1997the system Deep Blue developed by IBM to strike the chess world champion Garry Kasparov in six portions. Newell and Simon developed the general problem of Solver, a program, which with simple methods arbitrary problems can solve into the 1960er yearsshould, a project, which was stopped after nearly ten-year development duration finally.McCarthy suggested 1958, the entire human knowledge into a homogeneous, formal representational form, the predicate calculus 1. Stage to bring. The idea was to design theorem Beweiser which build symbolic expressions up,in order to discuss the knowledge of the world.

End of the 1960er years codeveloped Joseph wheat tree (* 1923) of with a relatively simple strategy the program ELIZA, in which the dialogue of a psychiatrist with a patient is simulated.The effect of the program was overwhelming. Wheat tree was surprised that one in relatively simple way humans can obtain the illusion of an inspired partner. In some areas the AI obtained successes, for example with strategy plays (chess, lady, etc.), during mathematical symbol processing,during the simulation of robots, when proving logical and mathematical sentences and finally with expert systems. In an expert system the rule-based knowledge of a certain field of activity is formally represented.
The system makes possible then with concrete questions, these rules automatically alsoto use in such combinations, which (of the human expert) were not explicitly seized before. The rules consulted for a certain problem solution can be spent then again also, i.e. the system can be result “to explain”. Individual knowledge elements can added, changed orare deleted; modern expert systems have in addition comfortable user interfaces.

One of the most well-known expert systems, at the beginning of the 1970er years of T. Shortliffe to the Stanford University developed system MYCIN for the support of diagnosis - and therapy decisions with blood infections and Meningitis,by an evaluation it was certified that its decisions are as good as an expert in the range concerned and better than a non--expert. However the system reacted, as to it data of an Cholera illness - as well known an intestine and noneBlood infection were entered -, with diagnostic and therapy suggestions for a blood infection, i.e., MYCIN did not recognize the borders of its authority. This Cliff and plateau effect is atypical not with expert systems, which are set high-specialized in a narrow field of knowledge.

Into the 1980er yearsthe role of a key technology was assigned to the AI, parallel to substantial progress at hard and software, in particular within the range of the expert systems. One expected various industrielle applications, perspectively also a separation “more monotonously” human work (and their costs) by AI-steered systems.After however many prognoses could not be kept, the industry and the research promotion reduced its commitment.

With the neural nets at the same time a new perspective of the AI stepped knocked against the light, among other things of work of Finnish engineer Teuvo Kohonen. In this range of the weak AI one solved from concepts from “intelligence” and analyzed themselves instead, on the basis of the neuro physiology, the information architecture of the human (/tierischen) brain. The modelling in form of artificial neural nets illustrated then, as from oneto be carried out knows very simple essential structure a complex pattern processing. Neuro computer science developed as scientific discipline for the investigation of these procedures.
It becomes clear that this kind of learning contrary to expert systems not on the derivation and use ofIs based to rules. From this it follows also that the special abilities of the human brain are not reducible on such a rule-based intelligence term within the animal realm. The effects of these insights on the AI research, in addition, on learning theory, didactics, the relationship to Consciousness and other areas are still discussed.

In the AI meanwhile numerous Subdisziplinen developed, so special languages and concepts for representation and use of knowledge, models to questions of revising barness, uncertainty and inaccuracy and machine learning procedures. The Fuzzy logicas the further form of the weak AI for instance with press controls was established.
Further successful applications of AI lie in the ranges of natural-language interfaces, sensor technology and robotics.

demarcation to other fields of computer science

as leaves itself an application, to thatone intelligence characteristics grants, from other applications differentiates? This question is not to be answered so easily, could one the point of view nevertheless hold that also already primitive operations such as addition, multiplication etc. intelligence from humans demand. One became an application neverthelessExecution of additions hardly as an application of artificial intelligence designate.

The problem lies here in the definition and demarcation of the intelligence term. A constructionalistic beginning for the solution of the problem consists of abstracting substantial intelligence characteristics of human intelligence andto measure then the abilities of a given application at these characteristics. This beginning brought numerous characteristics out, of which the following three are regarded at least as necessary conditions:

  1. The ability for the processing of arbitrary symbols (not only numbers).
  2. The structure oneinternal model of the outside world.
  3. The ability to an appropriate application of the knowledge.

Further characteristics and abilities, which however not when are necessarily judged, are logical concluding, Verallgemeinerung and specialization, use of natural speech etc. Altogether at present inclusive V1 existsto V3 of twelve relatively secured characteristics. It can be said now that an application must fulfill the V1 to V3, in order to be able to speak of an application of artificial intelligence. The more fulfilled, the more highly knows further characteristics applicationthe degree at intelligence, which application are realized evaluated. So far humans fulfill all of these characteristics, but did not succeed yet to develop an application at the same time the all 12 characteristics realizes.

Altogether is with described the hereto note constructionalistic beginning for the use of the intelligence term that this it defines numerous characteristics and makes so more understandable that on the other hand however the abundance of the characteristics is to be handled also relatively difficult.

philosophical aspects

the philosophical aspects thatBelong to AI problem to most extensive entire computer science. The answers, which are given on the central questions of this range, extend far into ontologische and epistemological topics inside, which already employed a thinking of humans in the beginnings of philosophy. Whosuch answers gives, must the conclusions from it also for humans and draw itself. Pretty often one would like to proceed in reverse and to transfer the answers, which one found before the development of artificial intelligence, to these. But was it shown as,artificial intelligence caused numerous researchers to it, problems like the relationship of subject and spirit, the origins of consciousness, the borders of the realization, the possibility of except-human intelligence etc. to regard in new light and partially again tooevaluate. See wants Machines Become Conscious?

see also

literature

  • Görz, Rollinger, Schneeberger (Hrsg.): Manual of artificial intelligence, 4. Edition 2003,Oldenbourg, ISBN 3486272128
  • Stuart Russell, Peter Norvig: Artificial Intelligence: A decaying Approach, 2. To edition, 2002, Prentice resound. (The current English-language standard work to the topic.)
  • Stuart Russell, Peter Norvig: Artificial intelligence: A modern beginning, August 2004,Pearson study, ISBN 3827370892 (German translation of the 2. Edition)
  • Karl stone book: Automat and humans. Berlin 1971
  • Dietrich Dörner: Structural drawing for a soul. Reinbek: Rowohlt, 2001. - ISBN 3-499-61193-7
  • Roger Penrose: Computer-think - the emperornew dresses or the debate around artificial intelligence, consciousness and the laws of nature, translation of the English expenditure for original “The Emperor's new Mind”, with a preface of Martin Gardner and a preface to the German expenditure of Dieter wall cutter, Heidelberg 1991
  • Roger Penrose: Shade of the spirit - ways to a new physics of consciousness, translation of the English expenditure for original “Shadows OF the Mind”, Heidelberg 1995
  • Howard Gardner: Thinking on the trace (AI as part of the interdisciplinary cognitive science), Stuttgart 1989, ISBN 3-608-93099-X
  • Marvin Minsky: Mentopolis, Stuttgart 1990, ISBN 3-608-93117-1
  • Douglas R. Hofstadter, Gödel, E, brook, an endless twisting volume, dtv, ISBN 3423300175
  • Rolf Pfeifer, Christian Scheier, Alex Riegler: Understanding Intelligence. Bradford Books. 2001. ISBN 026266125X
  • Michael Kary, Martin Mahner:How Would You Know if You Synthesized A Thinking Thing? Minds and Machines 12, 2002, 61-86.

Web on the left of

German

English

 

  > German to English > de.wikipedia.org (Machine translated into English)