History of AI

History of AI

The early work that is now generally recognized as AI was done in the period of 1943 to 1955. The first AI thoughts were formally put by men McCulloch and Walter Pitts (1943). Their idea of AI was based on three theories, firstly basic phsycology (the function of neurons in the brain), secondly formal analysis of propositional logic and third was Turing's theory of computation.

Later Donald Hebb in 1949 demonstrated simple updating rule for modifying the connection strengths between neurons. His rule now called Hebbian learning which is considered to be great influencial model in AI.

There were huge early day work that can be recognized as AI but Alan Turing who first articulated a complete vision of AI in his 1950 article named "Computing Machinery and Intelligence".

Real AI birth year is 1956 where in John McCarthy held workshop on automata theory, neural nets and study of intelligence where other researchers also presented their papers and they come out with new field in computer science called AI.

From 1952 to 1969 large amount of work was done with great success.

Newell and Simon's presented General Problem Solver (GPS) within the limited class of puzzles it could handle. It turned out that the order in which the program considered subgoals and possible actions was similar that in which humans approached the same problems. GPS was probably the first program which has "thinking humanly" approach.

Herbert Gelernter (1959) constructed the Geometry Theorem Prover which was capable of proving quite tricky mathematics theorem.

At MIT, in 1958 John McCarthy made major contributions to AI field :- development of HLL LISP which has became the dominant AI programing language.

In 1958, McCarthy published a paper entitled Programs with Common Sense, in which he described the Advice Taker, a hypothetical program that can be seen as the first complete AI system. Like the Logic Theorist and Geometry Theorem Prover. McCarthy's program was designed to use knowledge to search for solutions of problems.

The program was also designed so that it could accept new axioms in the normal course of operation, thereby allowing it to achieve competence in new areas without being reprogrammed. The Advice Taker thus embodied the central principles of knowledge representation and reasoning.

Early work building on the neural networks of McCulloch and Pitts also flourished. The work of Winogard and Cowan (1963) showed how a large number of elements could collectively represent an individual concept, with a corresponding increase in robustness and parallelism. Hebb's learning methods were enhanced by Bernie Widrow (Widrow and Hoff, 1960; Widro, 1962), who called his networks adalines, and by Frank Rosenblatt (1962) with his perceptrons. Rosenblatt proved the perceptron convergence theorem, showing that his learning algorithm could adjust the connection strengths of a perception to match any input data, provided such a match existed.

In 1965, Weizenbaum's ELIZA program appeared to conduct a serious conversation on any topic by basically borrowing and manipulating the sentences given by a human. None of the programs developed so far, had complex domain knowledge and were called 'weak' methods. Researchers realized that it was necessary to use more knowledge for more complicated, larger reasoning tasks.

The DENDRAL program was developed by Buchanan in 1969 and was based on these principles. It was a unique program that effectively used domain specific knowledge in problem solving. In the mid- 1970's, MYCIN, a program developed to diagnose blood infections. It used expert knowledge to diagnose illnesses and prescribe treatments. This program is also known as the first program, which addressed the problem of reasoning with uncertain or incomplete information.

Within a very short time a number of knowledge representation languages were developed such as predicate calculus, semantic networks, frames and objects. Some of them are based on mathematical logic such as PROLOG. Although PROLOG goes back to 1972, it did not attract wide spread attention until a more efficient version was introduced in 1979.

As the real, useful strong works on AI were put forward by researchers, AI emerged to be a big Industry.

In 1981, Japanese announced 5th generation project a 10-year plan to build intelligent computers running PROLOG. US also formed the Micro electronics and Computer Technology Corporation (MCC) for research in AI.

Overall the AI industry boomed from few million dollars in 1980 to billions of dollars in 1988. But soon after that AI industry had huge setback as many companies suffered as they failed to deliver on extra vagant promises.

In late 1970s more research were done by psychologists on neural networks which continued in 1980s.

In 1990s AI emerged as a science. In terms of methodology AI has finally come firmly under the scientific method. In recent years approaches based on Hidden Markov Models (HMMS) have come to dominate the AI field. This model is based on two aspects one is rigorous mathematical model theory and second is, these models are generated by a process of training on a large corpus real speech data.

Judea Pearl's (1988) Probabilistic Reasoning in Intelligent Systems led to a new acceptance of probability theory in AI. Later Bayesian network was invented which can represent uncertain knowledge along with reasoning support.

Judea Pearl, Eric Hovitz and David Hackerman in 1986 promoted the idea of normative expert systems that can act rationally according to the laws of decision theory. Similar but slow revolution have ocurred in robotics, computer vision and knowledge representation.

In 1987 a complete agent architecture called SOAR was work out by Allan Newell, John Laired and Paul Rosenbloom. Many such agents were developed to work in big environment "Internet". AI systems have become so common in web based applications that the "- bot" suffix has entered in everyday language.

AI technologies underlie many Internet tools, such as search engines, systems and website.

While developing complete agents it was realized that previously isolated subfields of AI need to reorganize when their results are to be tied together.