1. As an Engineering discipline concerned with the creation of intelligent machines.
2. As an empirical science concerned with the computational modeling of human intelligence. The second view is concerned with modern cognitive science. In this article we shall try to examine AI from this perspective.
The first step in the emergence of Al required the availability of usable computers. Some of the initial work in this area was done by Turing ( 1950) in his paper, 'Computing Machinery and Intelligence'.
But what is Artificial Intelligence? AI is actually about the design of intelligent agents. An agent is an entity that perceives and acts on its environment. It is also rational and its actions can be expected to achieve its goals when given information. This view of AI is a recent development in this field.
Initially, till the end 1980s, research in AI was concerned mainly with reasoning tasks, the inputs of which were provided by humans and the-outputs were also interpreted by them. Some of these task included mathematical theorem-proving systems, English question-answer systems etc. Today computer scientists are developing computer systems that-come very close to mimicking parts of human information processing and cognition.
The concept of AI is usually intertwined with cognitive psychology and neuroscience. All 3 of them build a platform for cognitive science. AI and cognitive psychology have a symbiotic relationship, where each benefits from the developments in the other.
This can be illustrated by the fact that in order to develop artificial ways of replicating human perception, memory, language, and thought, it becomes important for us to know how these processes are accomplished by human beings. Development of AI also increases the magnitude of our capabilities to understand human cognition. Another concept related to this area is the development of"intelligent" machines that model human thought. This perspective is caused "Computer Simulation".
Now that you have an idea of what AI is, let us go back to the beginnings of AI and see its historical background.
ARTIFICIAL INTELLIGENCE - HOW IT ALL BEGAN
Artificial Intelligence (AI) is the branch of computer science that attempts to write programs capable of higher mental processes that were traditionally associated with only humans (Gardner, 1985; Posner, 1986). Calculators are considered as the brain behind AI. They can be found in different forms across different civilizations. since the 6th century B.C. some of them are the abacus from China, a counting machine that counted with pebbles from Egypt and Greece etc. Some of the early digital calculators were the result of the inventions of Wilhelm Schickaed and Blaise Pascal. After these calculators, which could perform, simple mathematical calculations came the age of computers. The earliest computer was the invention of Charles Babbage.
Initially computers would also perform the same mathematical computations done by calculators, though with greater speed. Slowly these computers gave way to more complex and powerful systems, which could help people to solve the mathematical problems of commercial and industrial research.
In 1956, a very important date in cognitive psychology, scientists came together to consider the possibilities of developing computer programs that would behave intelligently, some of these scientists are John McCarthy, Marvin Minsky, Claude Shannon, Herbert Simon and Allen Newell. Since this conference, AI has grown tremendously. Today it touches the daily lives of many people, across the globe.
John Von Newmarn, a Hungarian mathematician, suggested that it might be possible to design a computer that mimicked the human brain in function as well structure. But 1 major difference between computers and humans is that while computers generally process information serially. humans do :so in parallel processing. One scientist who tried to overcome this was Daniel Hillis. He developed a "Connection Machine" which solves problems by breaking them down into smaller problems and then processing them in parallel. This machine has many processes that work on a single problem simultaneously.
Around this time, scientists were learning about the features of the brain and they were discouraged about the limits of the classical AI approach. According to some classical models, processing was viewed as a series of discrete operations i.e., one step must be completed before the system could go on to the next (Me Clelland, 1988). But cognitive activities involve parallel processing rather than serial processing. This brings us to the parallel-distributed processing approach (PDP) or Connectionism. This approach proposes that cognitive processes can be understood in terms of the networks that link together neuron-like units in the cerebral cortex of the brain. This outer-layer of the brain is responsible for cognitive processes. One important discovery in this area was the numerous connections among the neurons. This pattern of interconnections resembles many elaborate networks. The neural activity for a cognitive process is distributed throughout a section of the brain. Those who favored this approach, said that a model should be developed that resembled the important features of the brain. The most important characteristic of the PDP approach is that it is designed with the brain as a model instead of the computer (Palmer, 1987).
MACHINES AND MINDS
Supporters of AI believe that machines are not only capable of exactly replicating human cognition, but they can also carry out advanced intellectual processes. Therefore computers should be directly involved in our everyday decision-making. And then there are others, who say that human thinking is a purely human process and it can never be duplicated by AI programs. John Searle ( 1980) proposed that there are 2 forms of AI'weak' and 'strong'. 'Weak' AI can be used as a tool in the investigation of human cognition. 'Strong' AI is a properly programmed computer that has a 'mind' capable of understanding. Alan Turing had proposed one of the original mind-machine problems that has came to be known as the "Turing Test". Through this test it was concluded that for a computer to respond-like a human it must be able to understand and generate a response that effectively mimics some form of cognition. It also raises the question of whether one can find a difference in the functions performed by a human and a machine.
Another puzzle in the category of the mind-machine problem has been offered by Searle, popularly known as the "Chinese Room" puzzle. Searle through this concluded that though machines may perform functions, it does not understand the meaning of the "output". On the other hand, human minds have "intentionality". Even though machine thinking and human thinking may be indistinguishable, the two are not the same because of the intentions and understanding of the human thinker. This is the one that would discriminate bet ween a thinking person and a robot.
PERCEPTION AND AI
Perception is a very important part of human information processing. This function, which even children can easily perform, would be a complex problem for a computer. Naomi Weisstein (1973) has described how a simple perceptual task of finding a clock and reading the time, though as easy as child's play, would be an enormously complex task for a computer. Human perception is initiated by external signals of light, sound, molecular composition and pressure. These signals are detected by our sensory system and then sent as neural messages to the brain, which it understands. How would a machine initiate this perceptual mechanism? Perhaps a sensing capacity could be developed, as in computer recognition systems.
Computers recognize forms using a number of small "templates" that are systematically passed over each object in search of a match. The template is made up of 2 sensors-positive and negative. The procedure for the "perception" of a letter would require a huge computer memory (to store a template of each novel form of each letter) or it would fail to detect many valid forms of letters. In the area of letter-identification and form perception, AI has been ostracized because, there is no workable mechanism for attention. Unlike humans a machine sees a form as a whole pattern and finds it difficult to focus on local features. This issue is currently under active investigation.
The older AI alphanumeric recognition systems were based on the template concept.' In comparison the newer, neurally based computer model in actually capable of learning. Some of these computer implementations are capable of learning patterns, store them and recognize them later. One such program is called DYSTAL, which successfully acquires alphabetic letters and letter sequences and recognizes them even when only parts of the pattern are presented.
Some computer systems can perform like human experts.They are called 'expert systems'. An expert system is an artificial specialist that solves problems in its area of specialty. Such systems have been designed to solve problems in medicine, law, aerodynamics, chess etc. These systems follow rules and they can "think" about one issue only. In spite of this the "intellect" of computers is limited. Perceiving skills in humans improve with frequent experiences with objects and events. But can the same be said about computers? Humans get bored easily when they have to perform the same task continuously. But not the AI system. They can run endlessly without complaining. Thus skilled human perception of a repetitious act may be something a computer could do well. Computers have been successfully used to "see" faults and take simple decisions about the quality of a product. Visual detection can be done by using optics, which are far more sensitive than the human eye. Such inspections done by computers have been found to be faster, cheaper and more accurate. The capacity of a computer to "see" and recognize a limited but increasing number of complex sights is rapidly emerging. Now the task of the AI scientist and the cognitive scientist is to develop a machine that can store past information about the world and use that memory to abstract meaning from its precepts.
LANGUAGE AND AI
Language truly reflects human thought, perception, memory, problem solving, intelligence and learning. Since it is of importance for basic psychological principals, it has caught the interest of the AI scientists. The capacity for language and problem solving in computers can be illustrated through actual computer conversation. One of the first conversational computer programs called ELIZA was written by Joseph Weizebaum ( 1966). Through this program, a computer may be able to carry on a conversation with a human but its responses would tend to be stereotyped. It also lacks the conceptual base in terms of language and world knowledge that a human has. Human capacity for knowledge about feelings, tendencies, group dynamics etc. leads to understanding. This is where ELIZA falls short. Though a computer's conversation may sometimes fool people into thinking that a computer can have a real conversation with people and is able to "think", it can't fool people all the time. In spite of its limitless memory of words, and ability to produce meaningful sentences, it sometimes fails to fool people because of its lack of understanding of what language is all about. To overcome this, some very sophisticated "understanding" programs based on the conceptual base of language have been developed by Schank ( 1972, 1982), Shank and Hunter ( 1985), Anderson (1975), Anderson and Reiser ( 1985), Wilks (1973), and Winograd (1972, 1981, 1985). These systems have the capacity to analyze context of the discourse and meaning of the words and in some cases "world knowledge" (Winograd).
For an "intelligent" machine to understand language, it would have to make reasonable inferences about language processing just as humans do. But this may be difficult for a machine due to the ambiguous nature of language. To tackle this problem, an evolved program contains many systems and subsystems. Some of these systems are inferences, scripts, plans, goals, themes through which humans understand language etc., (Schank, 1981). Though such research may pave the way for automated companions that could become an indispensable part of everyday life, the outcome of all this would actually be a new kind of understanding of ourselves.
PROBLEM SOLVING AND AI
AI scientists have been interested in problem solving because it is roughly synonymous with thinking which is a unique human attribute. In addition to this, Al machines have the capability of problem solving and this has led to a lot of theories in this area. One of the earliest problem solving done by machines was calculation. In 1642, Pascal showed us how mathematical problems could be solved accurately and in less time by a mechanical calculator. Problem solving in the context of modern AI includes finding solutions to complex puzzles, proving theorems, playing games etc. There are 2 methods that can be used to solve a problem. They are algorithms and heuristics. Algorithms are commonly defined as procedures that guarantee a solution to a given problem. Heuristics are strategies that operate like a rule of thumb. Computers use heuristic search methods, in which strategy plays an important role.
Computers have the capacity for rapid and voluminous mathematics search and match activity, which may enable them to solve a problem faster than humans.
AI IN THE FORM OF ROBOTS
Robots incorporate everything that AI is all about the replication of pattern recognition, memory, language processing and problem solving. Robots are devices, which are capable of performing human tasks or behaving in a human manner. The growth of robotology can be attributed to the exploration of space and the need to develop highly sophisticated mechanical devices capable of performing specific tasks. The Mars landing is the result of one such need.
Some of the early prototypes of space robots were developed at the AI laboratory 1 of Stanford University. One very interesting robot developed was a mobile radio contrived vehicle called "Shakey" who had perceptual and problem solving capabilities. It. was equipped with a television camera, a range finder and a tactual sensor. They relayed sensory information to a computer, which then analyzed this information. The perceptual system reduced the picture to line images. The problem-solver was a type of theorem -proving program that allowed for the execution of simple tasks. A second version of Shakey was made with expanded memory and control system.
Some of the most advanced robots have been built by NASA. These robots have been designed to collect and analyze soil samples from other planets, repair space stations, carry out scientific experiment in environments where it may not be possible for humans to work. Today robots are used in the business community to perform laborious and dangerous functions.
AI AND EDUCATION
Cognitive scientists have tinkered long with the idea that perhaps computers could educate children as well as human tutors. One of the early computer tutoring systems called Computer Aided Instruction presented material, asked multiple-choice questions and gave further presentations depending on the student's answer (Dick and Cavey 1990). After this came the Intelligent CAl, which coached students as they worked complex problems such as writing a computer program. Some of these tutoring system.s are being regularly used in schools, industry and the military.
Though tutoring systems are a very popular use of AI in education, another application of AI in education is providing a simulated environment in which interactive learning can take place.
While developing a module of a tutoring system, a good understanding is required of how students learn so that the tutor's comments will prompt students to construct their overall understanding of the subject matter. An overly critical tutor may do more harm than good, therefore it is important to study how human tutors increase motivation and learning of the student. The findings may be used to formulate a more effective tutoring system.
Al AND NEUROSCIENCE
Cognitive neuroscience examines how the structure and function of the brain explain cognitive~ processes (Kosslyn and Koenig, 1992). This has been the base of the development of Artificial Information. AI scientists have used their knowledge of the brain's functioning in cognitive processes, to come up with machines which can perform similar functions, only more accurately and faster. Such machines are being used in almost every aspect of modern life. Neurological understanding of our cognitive process has helped these scientists to develop machines that are capable of very complex problem solving. In this endeavor, psychologists have provided rules and procedures for perception, storing information in memory, thinking etc., and computer scientists have used this information to write programs that would mimic these functions.
A recent development in this area is the building of computers that resemble a brain. They consist of layers of interconnected electronic surrogate neurons, whose organizational "hardware" mimics the "wetware" of the brain. Their programs mimic the functions of organic neural networks. These new computers are called "neural networks" which are more like humans than the earlier models. They are able to make generalizations and understand complex visual patterns. However the one thing that computers still do not have are emotions.
SUMMARY
Artificial Intelligence is concerned with computers that are programmed to think and act like humans. To achieve this goal, AI scientists have made use of the information provided by neuroscience and cognitive psychology. The calculator was one of the first products of AI. Slowly computers came into the picture that had the ability to solve complex problems. The new agenda for AI scientists at this point was to design a computer that would mimic the human brain in function and structure.
The result was advanced computers with perceptual, conversational and problem-solving capabilities.
Another interesting development in the area of AI is the robot. A robot is capable of pattern recognition. memory, language processing and problem solving. Today they are extensively used for space research. Artificial intelligence is also being used in the field of education in the form of tutoring systems called Computer Aided Instruction.
The major contribution to AI has been from cognitive neuroscience. AI has come this far only due to ample information provided about the structure and function of the brain in cognitive processes by psychologists and neuroscientists. This information has been used to develop machines that can successively perform some of the human cognitive processes.