Artificial intelligence

Artificial Intelligence (AI) is a research area of computer science with really no unanimous definition of the term. One definition is found in Ginsberg, which is that AI ​​is the enterprise of constructing an intelligent artifact or, more specifically, the project of building a physical-symbol system that can reliably pass the Turing test. Genesereth and Nilsson define AI as the study of intelligent behavior. Close to this definition, Luger defines AI as the branch of computer science that deals with the automation of intelligent behavior. A simple definition given by Elaine Rich and Kevin Knight is that AI is the study of how to make computers perform things that, at the moment, people do better. Russel and Norvig divides the numerous definitions of AI into four categories: Systems that think like humans, systems that act like humans; systems that think rationally; systems that act rationally. An interesting aspect of artificial intelligence is that it employs, and would be inconceivable without it, the concept of intelligent design.

Multiagent systems
Multiagent Systems (MAS) are a sub-area of ​​Distributed Artificial Intelligence and concentrate on the study of artificial agents that are autonomous and interact in an environment. Defined as systems composed of multiple interacting computational elements, known as agents. This environment can be closed or open. The multi-agent systems are typically distributed systems where several distinct components, each of which is an independent agent capable of solving problems, act together to form a coherent whole.

Chatterbots
Are programs that simulates virtual characters able to maintain a dialogue in natural language as if they were real humans. Examples of these systems are ELIZA and ALICE.

Intelligent tutoring systems
Systems used in education, essentially geared to supervised learning. Among the applications that are used, one can cite the areas of symbolic integration, troubleshooting electronics, mathematics based on axioms, medical diagnostics and gaming environments. Another application where this technique is applied is intelligent help systems.

Computer algebra systems
Computer algebra systems are examples of applications of AI in solving algebraic problems. Examples of such systems are Mathematica, Axiom, MatLab, Maple, Macsyma and its more modern version the Maxima.

Genetic algorithms
Genetic algorithms are search methods that use computer programming to find solutions to using methods inspired by biological evolution. Because they were inspired by the theory of evolution, some evolutionists claim them as evidence that microbe to man (or macroevolution) is possible. The traditional theory of genetic algorithms takes on a general level of description that they work by discovering, emphasizing, and recombining building blocks of good solutions in a highly parallel way.

Expert systems
Expert systems were one of the first successful applications in AI, and the Mycin was an example of successful expert system in the field of diagnostic medicine. The main components of an expert system is a knowledge base powered by a specialist, an inference engine and a working memory.

Symbolic and Numeric Processing
Symbolic processing involves the manipulation, within the computer, of symbols such as lists of words and characters. The languages LISP, Prolog and Smalltalk have been influencing AI concerning this aspect.

Natural Language Processing
A natural language is an ordinary language used as normal means of communication among people, such as Chinese, English, French, Japanese, Korean and Russian. The understanding of natural language is one the most difficult areas in AI, because it involves exchange of information between humans and computers. It involves the abilities of speech recognition, speech understanding, recognition of written characters, language understanding and language generation.

Pattern Matching
String-matching is the problem of locating all occurrences of a string s of length l, referred to as the pattern, in another string t of lenght m, referred to as the text. This search may differ in various aspects such as exact match and inexact match, ie allowing differences or gaps. The use of these techniques combined with dynamic programming is very useful in applications of sequence alignments. The comparison of sequences, particularly when combined with a search in databases containing molecular sequences became an essential part of modern molecular biology.

Fuzzy logic
Fuzzy logic is a branch of machine intelligence that deals with reasoning that is approximate rather than fixed and exact. In simple words, it helps computer paint gray, commonsense pictures of an uncertain world.

The limits of AI
Hubert Dreyfus in his provocative arguments in his 1972's book What computers can't do pointed out that it appeared that significant progress in cognitive simulation or AI was extremely unlikely, despite the optimism of the researchers in artificial intelligence. John R. Searle is another that emphasizes that computers actually are not aware of what they are doing. For Searle the Deep Blue computer, which defeated Gary Kasparov in 1996 does not "know" who is playing chess. To Searle,

The computer [Deep Blue] does not know that the symbols represent chess pieces and chess moves, because it does not know anything. (John R. Searle)

In the early 1950's Alan Turing, interested in the question whether machines could think, proposed a famous 'imitation game' now known as the Turing Test. The game is played with three people (a man, a woman and an interrogator) hidden from each other and the main purpose of the interrogator is to guess who is the man and who is the woman. Turing raises the question: if we could replace the man by a computer, it is possible that the interrogator could decide wrongly? Many objections to the test were discussed in the Turing's article itself and still others appeared. Faced with the question of whether it is possible that a computing machine can think, Feigenbaum and Feldman states that the answer is "no" if one defines thinking as an activity peculiarly and exclusively human or if one postulates that there is something in the essence of thinking which is inscrutable or mystical. However they states that if one admits that the question is to be answered by experiment and observation they think the answer to this question is "yes". However, even if a computer comes to pass the test proposed by Turing, there are still many who would argue that he could not think like a human and that the whole idea of strong AI is impossible to achieve. That is, even if a computer can behave as if it had the intelligence of a human being that does not mean he has the same understanding of a human being when behaving this way. Dembski also points out that even if artificial intelligence succeed in reducing intelligent agency to computation, cognitive scientists would still have the task of showing in what sense brain function is computational.

Ray Kurzweil, author of The Age of Spiritual Machines is an ardent advocate of Strong AI. He believes that in 30-40 years we will reach enough technology to build computer systems extremely powerful. For Kurzweil, we will be able to implement these powerful machines and neural networks with neurons functionally similar to those of humans. These neural networks, fed properly, Kurzweil hopes, will be able to be conscious. Kurzweil suggests that these machines are fed, for example, from a scanning of the human brain and reverse engineering the neuronal organization of the brain to feed these networks. Another alternative he proposes is to use genetic algorithms to develop a network with similar or greater capacity than the human brain in these networks. With this, we would have developed what Kurzweil calls the "Software of intelligence." Kurzweil gives hope that it may be possible that a recreation of the human brain has consciousness or at least behave as if:

"I don´t assume that a perfect or near-perfect recreation of a human brain would necessarily be conscious. But we can expect that it would exhibit the same subtle, complex behavior and abilities that we associate with humans. "

or:

"It may be that consciousness emerges from certain types of very complex self organitazing processes that take place in human brain".

In a close inspection of the Kurzweil's text, one can see that even he considers difficult or even impossible that we can understand how thoughts are formed and how we keep our memories. Precisely for this reason, he proposes that this information stored and organized in the human mind would be obtained through reverse engineering from a brain scanning. This is equivalent to someone making a pirated copy of an operating system. He gets a copy, but do not know how the software works. Of course, even if such copying is possible, there is no evidence that such a machine would have the same behavior as a human being, with desires, fears, dreams, anxieties, philosophical speculations and other characteristics of the human spirit. A second alternative suggested by Kurzweil to produce the "Software intelligence" is the use of genetic algorithms. What specifically is the difficulty of this approach? Genetic algorithms have a number of advantages such as parallelism, the solution space is wider, handle misunderstood search spaces among others. However, among the limitations of genetic algorithms are the problem of identifying the fitness function and of identifying a definition of representation for the problem. To create an appropriate fitness function would be necessary to understand exactly how a solution is better than another. That is, among which two solutions would be closer of consciousness than the other. Which one would have a better "software intelligence"? The behavior of the software can not be used as a parameter for the Kurzweil himself acknowledges that:

"My view is that consciousness, the seat of "personalness", is the ultimate reality, and is also scientifically impenetrable. In other words, there is no scientific test one can postulate that would definitively prove its existence in another entity".

Thus, to claim that the technique of genetic algorithms would lead to "Software intelligence" is misleading because it does not give the slightest clue how it would build a proper fitness function. And therein lies the real difficulty.