Cognitive science

from Wikipedia, the free encyclopedia
Illustration of the basic disciplines of cognitive science. Based on Miller, George A (2003). "The cognitive revolution: a historical perspective". TRENDS in Cognitive Sciences 7 .

Cognitive science is an interdisciplinary science for researching conscious and potentially conscious processes ( English science of the mind ).

The subject of cognitive science is conscious and unconscious experience, which is often localized between sensory and motor skills, as well as the processing of information in the context of human thinking and decision-making. These include B. Perception, thinking, judging, memory, learning and language. Your subject area is not limited to cognition , but also includes emotion , motivation and volition .

Cognitive science partially abstracts from whether cognition is examined in organic systems or living beings, or in artificial systems such as computers or robots, by considering cognitive processes as information processing. She works methodically on different levels:

  • the formation of theories, which is used to form hypotheses,
  • the cognitive modeling , cognitive performance using computer models to simulate and integrate new hypotheses in these models,
  • and the empirical level, which deals with the empirical verification of the models and the concrete implementation of cognitive performance.

Cognitive science is the result of interdisciplinary efforts between psychology , neuroscience , computer science / artificial intelligence , linguistics and philosophy , but also anthropology and sociology .

Development of cognitive science

History of Cognitive Science

Some associate the development of cognitive science with the idea of ​​a so-called “ cognitive turn ” (approx. 1940–1970). Until then had in the psychology and the mental philosophy of behaviorism played a major role. Behaviorism emerged as a response to the problems of introspection as a method of psychological research. Introspective reports on the mental inner workings were not externally verifiable for the scientists. Behaviorism drew the consequence that psychology had to limit itself to an investigation of behavior . In the philosophy of mind, for example, Gilbert Ryle went a step further and maintained that mental states were nothing more than behavioral dispositions.

In 1956 the Symposium on Information Theory took place at the Massachusetts Institute of Technology , in which the AI pioneers Allen Newell , Herbert A. Simon and Marvin Minsky as well as the linguist Noam Chomsky took part. Chomsky presented a harsh critique of behaviorism and introduced its influential transformational grammar . Newell and Simon presented the Logical Theorist , who for the first time was able to independently “prove” a mathematical theorem . Important forerunners of this development were the formulation of cybernetics by Norbert Wiener and the work of Alan Turing , who designed the Turing machine and developed the Turing test .

The cognitive science, which was constituted in the context of the developments described, was based on a central assumption that was called the “ computer model of the mind ”. This means the thesis that the brain is an information-processing system and basically works like a computer . The distinction between mind and brain can be understood analogously to the distinction between software and hardware . Just as software is determined by data structures and algorithms , the mind is determined by mental representations and computational processes. Just as the abstract description of the software is possible without directly examining the hardware, an abstract description of the mental abilities should be possible without directly examining the brain. And just as the existence of a software level can be easily reconciled with materialism , the mental level should also be embedded in a materialistic interpretation.

Current developments

The computer model of the mind has come under severe criticism over the past few decades. This criticism has essentially two sources: First, the description of the brain by cognitive neuroscience has developed rapidly. This can be seen, for example, in the increasing importance of imaging procedures , which make it implausible not to pay attention to the brain when exploring the mind. On the other hand, other successful approaches have developed, such as B. Connectionism and the modeling of neural networks . Artificial neural networks are programmed, among other things, to simulate the activities of groups of neurons . It is doubtful to what extent a distinction between software and hardware levels is still possible here.

Other alternative paradigms in cognitive science are e.g. As the dynamicism , Artificial Life (Artificial Life) and the materialized and-off cognitive science. According to dynamicism, the theory of dynamic systems provides a suitable model of cognitive behavior, since cognitive behavior always takes place in a temporal context and requires temporal coordination. It is postulated that this temporal aspect of cognition, which is neglected in the computer model of the mind, is essential. On the other hand, this approach questions the centrality of internal representation and symbol manipulation (see symbolism), since these concepts are not part of a dynamic explanation.

"Artificial life" is a term that contrasts with artificial intelligence: instead of solving abstract tasks (such as analyzing chess positions), which often seems difficult to us humans because of the sheer number of possible solutions, but computers are easy to solve , one should first understand how to cope with supposedly profane everyday problems. Many tasks that seem simple to us (such as running, recognizing friends and enemies, catching a ball ...) are currently not at all or only to a very limited extent capable of being mastered by computers or robots .

In turn, embodied and situated cognitive science assumes that cognition cannot be explained without reference to a specific body ( embodiment ) and a specific environment (situatedness). These demands result from the doubt that cognition is a process that takes place in a world of abstract symbolic representations, relatively independent of the exact sensory, motor and temporal events in the outside world. Well-known representatives of this view are Alva Noë , Susan Hurley, Evan Thompson , Francisco Varela and Kevin O'Regan. In the context of embodied and situated cognitive science, a link between the ideas of the phenomenology of Maurice Merleau-Ponty and Edmund Husserl and the classical analytical philosophy of mind is often sought.

These different currents presented (connectionism, dynamicism, artificial life, situatedness and embodiment) are often summarized under the catchphrase New AI , as they are e.g. T. overlap in their demands and assumptions. However, they cannot be regarded as congruent as they differ in many ways in terms of premises , consequences and applications, or even contradict one another.

The criticism of the computer model of the mind at times led to a general questioning of cognitive science. In the meantime, however, the waves have largely smoothed out. Cognitive scientists now also use neural networks themselves and are in close contact with cognitive neuroscience .

Philosophy of cognition

In cognitive science, topics are investigated that require consciousness or self-confidence in humans . For this purpose, individual aspects of consciousness such as perception, thoughts or memories are considered and generally referred to as mental states. “Higher” cognitive abilities such as learning, problem solving and speaking, in turn , require thinking - that is, mental states. It is therefore of great methodological importance for cognitive science to clarify what is meant by the talk of mental states. Associated with the computer model of the mind is a classic position in the philosophy of mind - functionalism .

Functionalism, developed by Hilary Putnam in the 1960s , claims that mental states are functional states. A functional state is specified by its causal role in a system . The concept of the functional state can be explained quite well using the example of simple machines : Let's imagine a candy machine . This throws out a candy for one euro. Now you can describe the machine with different states: There must be a state in which the machine ejects the candy without asking for further money. But there must also be states in which the machine still demands 1 euro or 50 cents to eject something. Each of these states of the machine is a functional state. It is specified by the fact that it reacts in a certain way to a certain input (here: 50 cents or 1 euro): it has a certain output (here: candy or not) and changes to another state.

1-band Turing machine

The decisive factor in this consideration is that the description of the functional state is independent of what and how the candy vending machine is actually built. If mental states were also functional states, it would also be irrelevant whether the functional state is realized in a brain or in a computer. This would also make clear the conditions that must be given so that a computer can have mental states: The computer only has to realize the same functional states. This also seems to be possible. The Turing machine , formulated as a mathematical model by Alan Turing in 1936 , can in principle realize any functional state.

Cognitive skills and cognitive architectures

People have a wide variety of cognitive abilities: memory , language , perception , problem solving , mental will , attention, and more. The aim of cognitive psychology is to explore the characteristics of these abilities and, as far as possible, to describe them in formal models . These models can then be implemented as cognitive architecture on a computer. The artificial intelligence (AI) has to realize the goal of cognitive skills in machines. However, in contrast to cognitive architectures, the artificial agents are also allowed to use strategies that are not used by humans.

Solve problem

“Problem solving” is the term used to refer to actions that are aimed at achieving a target state. Problem-solving processes are therefore something everyday, they are necessary for day planning , arithmetic, playing chess or planning a trip. Early on, the aim of artificial intelligence was to give machines the ability to solve problems.

A start and a target state are specified in the artificial intelligence. The task is to find the (or a) way to the goal. Here, there are basically two approaches: On the one hand, the program may try to blind them to their target to find by tried all different ways (so-called. Brute force method ), as for example in the depth-first search or breadth-first search is done. However, this approach reaches its limits very quickly, since the number of possible paths in NP-complete problems is so high that trying it out would exceed the computing capacity of the machine. In such a case, search algorithms using heuristics are necessary, such as the A * algorithm . Heuristics describe selection mechanisms that attempt to determine the most promising processes before they are carried out.

The first program that worked extensively with heuristics was the General Problem Solver (GPS) by Allen Newell and Herbert A. Simon . The GPS was able to find solutions such as the Towers of Hanoi game. The game consists of a number of discs of different sizes and three playing fields. At the start of the game, all discs are on the left-hand space. The goal is achieved when all discs are in the right field. Each disc may only lie on a larger disc and only one disc may be moved to either the left, the middle or the right place. Although the problem can be solved with an algorithm , people often solve this problem with heuristics, as the number of possible paths is growing rapidly.

The towers of Hanoi

Solving games like the Towers of Hanoi was a popular task in the early days of artificial intelligence. The reason for this is that only a very limited number of actions are possible here and there are no unpredictable events. The experimental verifiability of cognitive strategies was facilitated. Today one also dedicates oneself to complicated everyday tasks, such as the successful "execution" of a restaurant visit.

Cognitive architectures

The goal of a cognitive architecture is to summarize the various results of cognitive psychology in a comprehensive computer model. However, the results must be available in such a formalized form that they can be the basis of a computer program. By summarizing the individual results, on the one hand a comprehensive theory of cognition and on the other hand a commercially usable model should emerge. The three most successful cognitive architectures currently available are ACT-R ( Adaptive Control of Thought , ACT), SOAR and EPIC . With the PSI model , a further approach has been presented in recent years which, compared to the other architectures, is largely based on the current state of general psychology .

ACT-R is a production system with a number of modules . It consists of input and output modules, a production memory and a declarative memory. The target module defines which target should be pursued in the production system. In the production memory there are rules that determine which action is carried out when a selected goal is to be achieved, and which content must be in the working memory (or in different partitions of the working memory) so that the action can be carried out successfully. This "pattern matching" leads to the selection of a production rule and determines the action of the output module.

Cognitive architectures are characterized by the fulfillment of certain criteria, the Core Cognitive Criteria (CCC). These are:

A computer system that fulfills these characteristics is IBM's DeepQA .

Language and cognition

Mastery of the language is one of the outstanding cognitive abilities of humans. Having language skills is also a prerequisite for having some other cognitive skills. Without language, at least many thoughts could not be thought and many problems could not be solved. Language has therefore always played a central role in cognitive science. On the one hand, the question arises of how human language mastery is possible, and on the other, how one can make machines master the language.

Human language ability

How is it that people are usually able to learn languages? Up until the twentieth century, the prevailing opinion was that language acquisition can be explained by filtering out the rules of language in dialogue with other people. Such a position, called “ cognitivism ”, was represented by Jean Piaget . According to her, language skills are derived from general thinking skills. This theory was countered by Noam Chomsky for the first time with his " nativistic " position. Chomsky claims that humans are genetically endowed with a language organ that makes language acquisition possible in the first place. The language organ is located in the brain , but not as a firmly circumscribed neural region.

Noam Chomsky at the 2003 World Social Forum

Chomsky argues that language acquisition cannot be explained by a cognitivistic approach. The linguistic input of fellow human beings is insufficient to determine the rules of correct speaking. On the one hand, the spoken language is very often ungrammatical, so the input is deficient. On the other hand, the input allows grammatical errors in learning children, which they do not actually make. Chomsky concludes that there must be innate language knowledge that can be used in language acquisition. This innate knowledge is especially grammatical knowledge, all people are given a universal grammar from birth.

Chomsky's hypotheses were heavily criticized in the scientific debate known as Linguistics Wars of the 1960s and 1970s: his syntax-oriented interpretative semantics from George Lakoff and his universal grammar from Benjamin Whorf's so-called linguistic relativity theory .

Since the 1980s, research has increasingly turned back to concepts that - like Piaget - focus on socialization in language acquisition. Chomsky's approach is - like the entire traditional "head philosophy" - challenged in constructivist concepts and through neurobiological models:

According to Humberto Maturana and Francisco Varela - see also: The Tree of Knowledge (El árbol del conocimiento 1984) - the brain is not constructed like an input / output model, but has - through a network of one hundred billion inter- neurons , the millions of Connect motor and sensory nerve cells with one another - the ability for intensive parallel processing . A representative idea with an image of a concept in the brain is hardly tenable for Maturana and Varela, since hundreds of neurons from other parts of the nervous system converge at the switching points with a variety of effects and lead to overlapping. The nervous system does not work with representations of an independent outside world. Words as designations of objects or situations in the world do not do justice to the fact of structural coupling, rather they are ontologically established coordination of behavior . According to Maturana and Varela, language does not arise in a uniform design (it is not part of the brain), but is the variable communicative behavior learned through coordination of actions (language is part of the milieu that is called the "realm of language": our common " Being-in-the-language [-] is what we experience as consciousness or as 'our spirit' and 'our I' ".)

Dialogue and expert systems

The attempt to equip machines with language capabilities is often reflected in dialogue systems . A dialog system is usually a computer program that can be used to chat using the keyboard. One of the first successful dialogue systems was ELIZA by Joseph Weizenbaum from 1966. ELIZA simulates a psychotherapist. Through the skillful use of phrases such as “Tell me more about X” or “Think about X often”, ELIZA was able to deceive test persons about their non-human existence for a long time. Some test persons even felt that they were understood so well that they wanted to talk to ELIZA about their problems privately beyond the test situation. However, if you ask ELIZA questions that do not fit into the context of the therapy situation, ELIZA is unable to provide any reasonable answers.

Joseph Weizenbaum (1923–2008), the inventor of ELIZA

Expert systems , which meanwhile also have numerous commercial applications, are related to dialogue systems . Expert systems try to store the knowledge of human experts and make it available to the user. Applications are, for example, automatic medical or technology experts. These experts require a functioning knowledge representation through which the program has the knowledge. In a comprehensive knowledge representation, the material must be structured in a favorable way, so that the necessary knowledge can always be accessed, that the relationships between the knowledge elements are clear and that the content can be overlooked by the developer and expanded if necessary.

The Turing test

The fascination of dialogue systems is closely related to a thought experiment formulated by the computer pioneer Alan Turing in 1950. Turing was looking for a clear criterion for deciding when computers can be considered intelligent. His answer was the famous Turing test: a person enters into dialogue with a computer - using a screen and keyboard. The computer can be regarded as intelligent precisely when it is difficult for a person to decide whether it is a question of a dialogue with a person or with a computer program.

Today's dialogue systems are still a long way from passing the Turing test. This is not surprising when you consider what a program would have to be able to do to pass it. It should be able to explain jokes, understand allusions and irony , and formulate questions and answers that are appropriate to the context. There is now the $ 100,000 Loebner Prize for the developer of the first program to pass the Turing test.

Much criticism has been made of the Turing test. The best known is probably John Searle's Chinese Zimmer argument, which is supposed to show that passing the Turing test is not sufficient for understanding language. Imagine being in a huge library. From outside you are handed in sheets of paper with Chinese characters that you cannot understand. Since only sequences of Chinese characters are recorded in the books in the library, you can now look for the character sequences on the sheets. In the book, each character sequence is assigned a different character sequence, which is finally written on the sheet and then given out again. This procedure makes it appear to an outside Chinese that he is talking to another person who understands Chinese. You don't understand Chinese yourself and the library doesn't understand Chinese either. So a system could pass the Turing test without even understanding what is being said.

The connectionism

Simplified representation of an artificial neural network

In cognitive science, the development of connectionism has brought about major changes. While in classical artificial intelligence - according to the computer model of the mind - cognitive abilities were simulated with a symbolic programming language , in connectionism artificial neural networks are used . An artificial neural network is an interconnection of simple units, the so-called artificial neurons . The neurons can pass on their activities to the neighboring neurons. As a result, complicated arousal patterns can arise with a given input, which in turn produce an output.

The concept of neural networks was developed in 1943 by Warren McCulloch and Walter Pitts . In 1949, the psychologist Donald O. Hebb developed Hebb's learning rule , which can be integrated into the concept of neural networks. According to Hebb, learning can be described by weighting the individual connections between the neurons. Learning takes place by changing the weights between the neurons. Despite this early development towards a model of learning neural networks, cognitive science remained limited to the symbol processing approach (GOFAI) for a long time.

It is only since the 1980s that cognitive science has increasingly used neural networks. This is particularly due to the fact that neural networks are able to perform tasks in which the symbol-processing approach has remained quite unsuccessful. Such tasks include, for example, pattern recognition or movement . This development is also of theoretical importance: Connectionism no longer recognizes the distinction between software and hardware, which is so important for classical cognitive science.

Cognitive Science at Universities

In the USA, but also in Great Britain, Australia and the Netherlands, cognitive science is a widespread and recognized subject. There are influential institutes at Rutgers University , Tufts University , the University of California, San Diego and the University of California, Berkeley .

In Germany, however, cognitive science is not yet very widespread as a degree. The University of Osnabrück has its own cognitive science institute with a bachelor's , master's and doctoral program ; the University of Tübingen has had a bachelor's and master's degree in cognitive science since the 2009/10 winter semester, offered by the Faculty of Mathematics and Natural Sciences. The Technical University of Darmstadt has been offering the "Cognitive Science" course since the 2019/20 winter semester . Cognitive science can be studied as a minor subject at the Albert Ludwig University of Freiburg and the University of Potsdam . Since the winter semester 2012/2013 an M.Sc. Course offered. Since the winter semester 2013/2014, the English-language course Cognitive Science (M.Sc.) has been offered at the TU Kaiserslautern . Related subjects are the bachelor's degree in cognitive computer science at Bielefeld University , the bachelor's degree “Philosophy - Neurosciences - Cognition” at Otto von Guericke University Magdeburg and MEi: CogSci, the joint degree “ Middle European interdisciplinary master program in Cognitive Science ”, which the universities in Vienna, Bratislava, Budapest and Ljubljana offer together. The University of Duisburg-Essen offers a bachelor's and master's degree in "Applied Cognitive and Media Studies". At the Technical University of Chemnitz there has been a bachelor's and master's degree in Sensor Technology and Cognitive Psychology since the 2009/10 winter semester, which focuses on technical sensor technology, human perception and natural and artificial cognitive systems.

See also

Portal: Mind and Brain  - Overview of Wikipedia content on Mind and Brain



  • John R. Anderson : Cognitive Psychology. An introduction . Spectrum of Science, Heidelberg 1988, ISBN 3-922508-19-7 . A well-founded introduction, but with little reference to neuroscience.
  • Howard Gardner : On the trail of thinking. The way of cognitive science . Klett-Cotta, Stuttgart 1989 a. ö., ISBN 3-608-93099-X , ISBN 3-608-95866-5 . Classical presentation of the history of cognitive science.
  • Manuela Lenzen : Natural and Artificial Intelligence. Introduction to Cognitive Science . Campus, Frankfurt am Main u. a. 2002, ISBN 3-593-37033-6 . Short, lay-friendly introduction.
  • Rolf Pfeiffer and Christian Scheier : Understanding Intelligence . MIT Press, Cambridge (Mass.) 1999, ISBN 0-262-16181-8 . Presentation of modern approaches in cognitive research.
  • Paul Thagard : Cognitive Science. A textbook . Klett-Cotta, Stuttgart 1999, ISBN 3-608-91919-8 . Also a lay-friendly introduction, focusing on philosophical and methodological aspects.
  • Max Urchs: machine - body - mind. An Introduction to Cognitive Science . Vittorio Klostermann, Frankfurt am Main 2002, ISBN 3-465-03196-2 . Comprehensive but understandable introduction from a mathematician and philosopher.
  • Francisco J. Varela : Cognitive Science, Cognitive Technique. A sketch of current perspectives . Suhrkamp, ​​Frankfurt am Main 1990, ISBN 3-518-28482-7 . Describes the biologically oriented, but not so much the classic cognitive science based on the computer metaphor, in an amateur-friendly way.

Text collections:


  • Robert A. Wilson , Frank C. Keil (Eds.): The MIT Encyclopedia of the Cognitive Sciences . MIT Press, Cambridge (Mass.) Et al. a. 2001, ISBN 0-262-73144-4 . English-language standard work.
  • Gerhard Strube et al. (Ed.) Dictionary of Cognitive Science . Klett-Cotta, Stuttgart 1996, ISBN 3-608-91705-5 . As CD-Rom: Klett-Cotta, Stuttgart 2001, ISBN 3-608-94167-3 .

Individual topics:

  • Ansgar Beckermann : Analytical Introduction to the Philosophy of Mind . 2nd Edition. De Gruyter, Berlin a. a. 2001, ISBN 3-11-017065-5 . Very dense introduction to the philosophy of mind.
  • Rainer Dietrich: Psycholinguistics . Metzler, Stuttgart 2002, ISBN 3-476-10342-0 . Layperson-friendly introduction to the cognitive science aspects of linguistics, but without neurolinguistics.
  • E. Bruce Goldstein: Cognitive Psychology. Connecting Mind, Research and Everyday Experience . Thomson Wadsworth, Belmont (Calif.) Et al. a. 2004 and other , ISBN 0-534-57726-1 . One of the newest and most widely used textbooks in cognitive psychology.
  • Klaus Mainzer : AI - Artificial Intelligence. Basics of intelligent systems . Primus, Darmstadt 2003, ISBN 3-89678-454-4 . Introduction to AI written by a science theorist. Therefore also understandable for non-IT specialists.
  • Horst M. Müller: Psycholinguistics - Neurolinguistics. The processing of language in the brain . UTB, Paderborn 2013, ISBN 978-3-8252-3647-2 .

Web links

Wikibooks: Brain and Language  - Learning and Teaching Materials
Thematic introductions
Institutes and research groups
Databases and collections of links on articles and researchers
List of international institutes offering Cognitive Science degrees

Individual evidence

  1. See Margaret Boden: Mind as Machine. A History of Cognitive Science , Oxford University Press, Oxford 2006, pp. 10ff.
  2. Chris Eliasmith: How to Build a Brain: A Neural Architecture for Biological Cognition . Oxford University Press, 2013, ISBN 978-0-19-979454-6 .
  3. Heinz von Foerster and others: Introduction to Constructivism . Publications of the Carl-Friedrich-von-Siemens-Stiftung, 5; Munich, Zurich: Piper-TB, 2006.
  4. Humberto R. Maturana and Francesco J. Varela: The tree of knowledge. The biological roots of human knowledge . Frankfurt 2010, p. 175ff.
  5. ^ Maturana and Varela, 2010, p. 226.
  6. Maturana and Varela, 2010, p. 251