Approaches Artificial intelligence

1 approaches

1.1 cybernetics , brain simulation
1.2 symbolic

1.2.1 cognitive simulation
1.2.2 logic-based
1.2.3 anti-logic or scruffy
1.2.4 knowledge-based


1.3 sub-symbolic

1.3.1 embodied intelligence
1.3.2 computational intelligence , soft computing


1.4 statistical
1.5 integrating approaches

approaches

there no established unifying theory or paradigm guides ai research. researchers disagree many issues. few of long standing questions have remained unanswered these: should artificial intelligence simulate natural intelligence studying psychology or neurology? or human biology irrelevant ai research bird biology aeronautical engineering? can intelligent behavior described using simple, elegant principles (such logic or optimization)? or require solving large number of unrelated problems? can intelligence reproduced using high-level symbols, similar words , ideas? or require sub-symbolic processing? john haugeland, coined term gofai (good old-fashioned artificial intelligence), proposed ai should more referred synthetic intelligence, term has since been adopted non-gofai researchers.

stuart shapiro divides ai research 3 approaches, calls computational psychology, computational philosophy, , computer science. computational psychology used make computer programs mimic human behavior. computational philosophy, used develop adaptive, free-flowing computer mind. implementing computer science serves goal of creating computers can perform tasks people accomplish. together, humanesque behavior, mind, , actions make artificial intelligence.

cybernetics , brain simulation

in 1940s , 1950s, number of researchers explored connection between neurology, information theory, , cybernetics. of them built machines used electronic networks exhibit rudimentary intelligence, such w. grey walter s turtles , johns hopkins beast. many of these researchers gathered meetings of teleological society @ princeton university , ratio club in england. 1960, approach largely abandoned, although elements of revived in 1980s.

symbolic

when access digital computers became possible in middle 1950s, ai research began explore possibility human intelligence reduced symbol manipulation. research centered in 3 institutions: carnegie mellon university, stanford , mit, , each 1 developed own style of research. john haugeland named these approaches ai old fashioned ai or gofai . during 1960s, symbolic approaches had achieved great success @ simulating high-level thinking in small demonstration programs. approaches based on cybernetics or neural networks abandoned or pushed background. researchers in 1960s , 1970s convinced symbolic approaches succeed in creating machine artificial general intelligence , considered goal of field.

cognitive simulation

economist herbert simon , allen newell studied human problem-solving skills , attempted formalize them, , work laid foundations of field of artificial intelligence, cognitive science, operations research , management science. research team used results of psychological experiments develop programs simulated techniques people used solve problems. tradition, centered @ carnegie mellon university culminate in development of soar architecture in middle 1980s.

logic-based

unlike newell , simon, john mccarthy felt machines did not need simulate human thought, should instead try find essence of abstract reasoning , problem solving, regardless of whether people used same algorithms. laboratory @ stanford (sail) focused on using formal logic solve wide variety of problems, including knowledge representation, planning , learning. logic focus of work @ university of edinburgh , elsewhere in europe led development of programming language prolog , science of logic programming.

anti-logic or scruffy

researchers @ mit (such marvin minsky , seymour papert) found solving difficult problems in vision , natural language processing required ad-hoc solutions – argued there no simple , general principle (like logic) capture aspects of intelligent behavior. roger schank described anti-logic approaches scruffy (as opposed neat paradigms @ cmu , stanford). commonsense knowledge bases (such doug lenat s cyc) example of scruffy ai, since must built hand, 1 complicated concept @ time.

knowledge-based

when computers large memories became available around 1970, researchers 3 traditions began build knowledge ai applications. knowledge revolution led development , deployment of expert systems (introduced edward feigenbaum), first successful form of ai software. knowledge revolution driven realization enormous amounts of knowledge required many simple ai applications.

sub-symbolic

by 1980s progress in symbolic ai seemed stall , many believed symbolic systems never able imitate processes of human cognition, perception, robotics, learning , pattern recognition. number of researchers began sub-symbolic approaches specific ai problems. sub-symbolic methods manage approach intelligence without specific representations of knowledge.

embodied intelligence

this includes embodied, situated, behavior-based, , nouvelle ai. researchers related field of robotics, such rodney brooks, rejected symbolic ai , focused on basic engineering problems allow robots move , survive. work revived non-symbolic viewpoint of cybernetics researchers of 1950s , reintroduced use of control theory in ai. coincided development of embodied mind thesis in related field of cognitive science: idea aspects of body (such movement, perception , visualization) required higher intelligence.

computational intelligence , soft computing

interest in neural networks , connectionism revived david rumelhart , others in middle of 1980s. neural networks example of soft computing --- solutions problems cannot solved complete logical certainty, , approximate solution sufficient. other soft computing approaches ai include fuzzy systems, evolutionary computation , many statistical tools. application of soft computing ai studied collectively emerging discipline of computational intelligence.

statistical

in 1990s, ai researchers developed sophisticated mathematical tools solve specific subproblems. these tools scientific, in sense results both measurable , verifiable, , have been responsible many of ai s recent successes. shared mathematical language has permitted high level of collaboration more established fields (like mathematics, economics or operations research). stuart russell , peter norvig describe movement nothing less revolution , victory of neats . critics argue these techniques (with few exceptions) focused on particular problems , have failed address long-term goal of general intelligence. there ongoing debate relevance , validity of statistical approaches in ai, exemplified in part exchanges between peter norvig , noam chomsky.

integrating approaches

intelligent agent paradigm
an intelligent agent system perceives environment , takes actions maximize chances of success. simplest intelligent agents programs solve specific problems. more complicated agents include human beings , organizations of human beings (such firms). paradigm gives researchers license study isolated problems , find solutions both verifiable , useful, without agreeing on 1 single approach. agent solves specific problem can use approach works – agents symbolic , logical, sub-symbolic neural networks , others may use new approaches. paradigm gives researchers common language communicate other fields—such decision theory , economics—that use concepts of abstract agents. intelligent agent paradigm became accepted during 1990s.


agent architectures , cognitive architectures
researchers have designed systems build intelligent systems out of interacting intelligent agents in multi-agent system. system both symbolic , sub-symbolic components hybrid intelligent system, , study of such systems artificial intelligence systems integration. hierarchical control system provides bridge between sub-symbolic ai @ lowest, reactive levels , traditional symbolic ai @ highest levels, relaxed time constraints permit planning , world modelling. rodney brooks subsumption architecture proposal such hierarchical system.