Artificial intelligence (AI) A branch of computer science that was pursued with great optimism in the 1960s and 70S, in an attempt to make computers think more like human beings. It analyzes data and draw conclusions in a way that makes them appear to be “intelligent.” Now of course, computers can’t think or do anything without being told what to do. So AI programs use complex formulas which attempt to arrive at an answer in a method similar to how a human might do it.
Some AI programs attempt to simulate the way humans think, while others try to mimic an expert with years of experience. Two AI categories that deal with the way humans think are fuzzy logic and neural networks. Fuzzy logic programs take into account the fact that input is not always precise sometimes input requires a judgment.
For example, an average traffic light turns red or green for a given period of time, regardless of the traffic flow. However, if the traffic light contained an AI computer using fuzzy logic, it could judge when to change the light based on the flow of traffic. The AI computer would not make a decision based on exactly how many cars were passing through the intersection, but rather, on whether that traffic was considered “heavy” or “light.” Because traffic flow varies, the AI computer would examine the changing number of cars and make a judgment on when to change the light.
Neural networks are almost the complete opposite of fuzzy logic systems. Neural networks, modeled after the neurons and synapses of the human brain, are programmed to reach a “best guess” conclusion based on concrete data (data that doesn’t change). By comparing all the data it has, a neural network can draw a fairly accurate prediction of the future. Neural networks have been used in studying economic and financial trends, as well as the stock market.
But the most popular AI programs today are not those that simulate human thought patterns, but those that mimic “experts.” An expert system contains all the “rules of thumb” normally acquired through years of experience. With its vast database of knowledge, an expert system can analyze a present situation and offer a possible solution based on all the “rules” that it has learned. AI is still in its infancy, as scientists learn more and more about the way humans think. AI programs are useful in many ways, from controlling the temperature in a building (making a judgment as to what’s comfortable) to predicting the stock market. In the future, AI programs could be used to pilot a plane, control a car, or design the perfect city. Robots with complex AI programs could perform specific tasks, such as housekeeping, farming, or policing a city-but they will never be like humans.
Computers are great at analyzing data at incredible speeds, but even with advances in AI technology, computers will never be terrifically good at independent judgment or handling unexpected or new situations.
Typical problems tackled in AI departments are COMPUTER VISION, NATURAL LANGUAGE understanding and advanced ROBOTICS the meager rate of progress led to a more realistic understanding that these problems are vastly more difficult than was then thought.
Artificially intelligent software can “learn” in a rudimentary fashion, adjusting the solutions it comes up with on the basis of past experience. But no expert system can begin to imitate the flexible intelligence of human beings.
Alan Turing proposed a strict test for artificial intelligence years ago. He said, “A machine has artificial intelligence when there is no discernible difference between the conversation generated by the machine and that of an intelligent person.” His test goes like this: In Room A you sit a person at a computer. In Room B you sit another person at a computer, plus you have a separate computer that claims to be intelligent, controlling itself. Using her computer, the person in Room A converses with the person and with the “intelligent” computer in Room B, one at a time, without knowing which one she is “talking” to. The person in Room A could type “How’s the weather” or “What is the speed of light in a vacuum?” or “What are you going to do after breakfast?” and the person or computer in Room B would send back the answer. If the person in Room A can’t tell whether she’s talking to a person using a computer or to the self-controlled computer, then the computer that is controlling itself is artificially intelligent.
As Arthur Naiman says of this test, “No computer in existence today can even dream of passing the Turing test-but then again, neither can most newscasters, corporate executives, or four-star generals.”
The programming languages LISP and PROLOG and techniques such as EXPERT SYSTEMS, NEURAL NETWORKS and case-based reasoning all emerged from university AI research. The term AI became rather unfashionable (especially with funding bodies) and tends to be replaced nowadays by less hubristic titles such as Knowledge Engineering. A similar tendency is to sneak elements of AI into ordinary software products so that the user is aware only of improved performance; for example a word processor that can generate an automatic summary of a document.
Artificial Intelligence (AI) definition may be examined more closely by considering the field from three points of view: computational psychology, computational philosophy, and machine intelligence.
We’ll be covering the following topics in this tutorial:
The goal of computational psychology is to understand human intelligent behavior by creating computer programs that behave in the same way that people do. For this goal it is important that the algorithm expressed by the program be the same algorithm that people actually use, and that the data structures used by the program be the same data structures used by the human mind. The program should do quickly what people do quickly, should do more slowly what people have difficulty doing, and should even tend to make mistakes where people tend to make mistakes.
The goal of computational philosophy is to form a computational understanding of human-level intelligent behavior, without being restricted to the algorithms and data structures that the human mind actually does use. By “computational understanding” is meant a model that is expressed as a procedure that is implementable on a computer. By “human-level intelligent behavior” is meant behavior that, when engaged in by people, is commonly taken as being part of human intelligent cognitive behavior, even though the implemented model performs some tasks better than any person would.
The goal of machine intelligence is to expand the frontier of what we know how to program. This goal led to one of the oldest definitions of AI: the attempt to program computers to do what, until recently, only people could do. This is a perpetually self-defeating goal in that, as soon as a task is conquered, it no longer falls within the domain of AI. Two examples are computer algebra and response to database queries. Thus, AI is left with only its failures; its successes become other areas of computer science.
The goal of machine intelligence differs from that of computational psychology and computational philosophy in being task-oriented rather than oriented toward the understanding of general intelligent behavior. A machine intelligence approach to a task is to use any technique that helps accomplish the task, even if the technique is not used by humans and would probably not be used by generally intelligent entities.