High-Tech Aplications of Cognition: Artificial Intelligence (AI), Expert Systems, Artificial Neural Networks (ANNs) and Robotics
Artificial Intelligence (AI)
Can a computer really think like a human being? Yes and no, As you'll see. Although there are important differences between the ways computers and human brains process information, amazing progress has been made in the field of artificial intelligence since the term was first used by researcher John McCarthy in 1956. Artificial intelligence (AI) refers to the programming of computer systems to simulate human thinking in solving problems and in making judgements and decisions. The first successful effort to program computers to mimic human thinking was made by Allen Newell and Herbert A. Simon in 1972. They developed programs that could play chess as well as human expert could, although not as well as master players. Eventually, AI experts at IBM created a computer called Deep Blue, wich defeated long-time world chess champion Garry Kasparov in 1997. (In 2003, a match between Kasparov and a smaller, slower version of Deep Blue known as Deep Junior ended in a draw.)
Some experts have stated that chess-playing computers such as Deep Blue could not really think on their own. These experts insist that even the most sophisticated "thinking" of computers is not really analogous to human cognition, because, unlike humans, these machines are merely following a set of rules, no matter how complex those rules may be. Stunned by his defeat, Kasparov disagreed with Deep Blue's critics. As the match proceeded, he felt that the computer was showing signs of human-like cognitive ability in the form of strategic understanding. Somewhere along the way, the tactics (specific rules for playing chess) that had been programmed into Deep Blue had apparently been transformed into strategy―the formulation of an overall game plan.
Decades before Deep Blue's stunning defeat of the reigning world chess champion, Newell and Simon had opened the door for computer systems able to do far more than play chess like an expert. Today's expert systems "perform a substantial number of human tasks at a professional level" (Simon, 1995, p. 507). Expert systems are computer programs designed to carry out highly specific funtions within a limited domain. One of the first medical expert systems was MYCIN, a computerized diagnostician in the area of blood diseases and meningitis. Other expert systems perform a range of funtions in medicine, space technology, military defense, weather prediction, and a variety of other fields.
Expert systems operate in well-defined domains; that is, such a system's knowledge base, or store of information, is not inordinately large and is limited to a specific area of expertise (Hendler, 1994). Outside its area of expertise, an expert system not funtion. Expert systems, then, are useful only as assistants to humans; they cannot stand alone. For example, expert systems can be used by physicians to confirm diagnoses or to suggest diagnoses doctors haven't thought of (Brunetti et al., 2002). Yet, these systems far surpass the human brain in their ability to retrieve massive amounts of stored data and to use that data to make decisions based on specific facts and rules that have been preprogrammed into their software. In addition, because they run on computers, expert sistems are vastly superior to humans in carrying out complex mathematical operations at lighting speed.
Moreover, the idea that expert systems can't stand alone is being challenged. For example, some studies show that psychotherapy clients can be helped just as effectively by interacting with expert systems as in face-to-face sessions with a human therapist (Taylor & Luce, 2003). Remember, though, that any expert system relies on the accumulated knowledge of human experts. Thus, it is impossible for computers to totally replace human professionals.
Artificial Neural Networks
Scientists are devising even more sophisticated computer systems, based on their understanding on how neurons in certain parts of the brain are connected and how those connections develop (Buonomano & Merzenich, 1995; Hinton et al., 1995). Computer systems that are intended to mimic the human brain are called Artificial Neural Networks (ANNs). Like synaptic connections in the brain, the connections in an ANN can be strengthened or weakened as a result of experience. Many of the newer expert systems utilize ANNs, making them not only able to process information like human experts but also to learn from experience just as those experts do (Brunetti et al., 2002). However, even some of the simplest cognitive tasks that people accomplish without conscious effort can be very difficult to program into a computer system. For example, ANNs that can learn how to distinguish between moving beings, such as humans, horses, cows, and so on, have only recently been developed (Tabb et al., 2002).
Understanding what people say to us seems to be a fairly simple task, one that we take for granted. However, comprehension of natural human language is by far the biggest challenge for ANNs designers. In one of the pioneering studies in this field, computer scientist Alex Waibel and colleagues developed a system:
"programmed to modify itself according to whatever signals come into the system... the speech recognizer actually learns how to identify sounds and words" (Peterson, 1993, p. 254).Voice recognition systems are already used by banks and credit card companies and in other commercial settings. But, unlike humans, such computer systems cannot understand the subtleties of language―tone of voice, accompanying nonverval behavior, or even level of politeness (Peterson, 1993).
Many other aspects of language processing also are extremely difficult for computers to manage. For example, what kind of scene comes to mind when you hear the word "majestic"? Perhaps you see a range of snow-capped mountains. At present, computer scientists are working to develop programs that can enable computers to retrieve images on the basis of such vague, abstract cues (Kuroda, 2002).
It's important to note here that a simple "if-then" program would allow a computer to call up an image of a mountain range when the word "majestic" was keyed in. So, the research with ANNs is not just to use a computer to bring about a particular result, but to be able to program the computer to produce the result in the same way that the human mind does. Perhaps you visualize mountains when you hear the word "majestic" because this word is linked to height as a scene feature. The concept of height is, in turn, associated with mountains. However, skycrapers are tall as well, so why don't you see them in your mind's eye when you hear "majestic"? Perhaps some people do. If so, why does one person see skycrapers while another sees mountains? Is there a particular experience or piece of knowledge that leads to this difference? These are the kinds of cognitive pathways that networks designers attempt to identify in human thinking and duplicate in computer programs. In so doing, these scientists not only program computers to accomplish tasks that seemed nearly impossible just a few short decades ago but, at the same time, gain insight into human cognition.
Do you recall the article about artificial limbs? One area in which human cognition has been applied to develop technological marvels is Robotics―the science of automating human and animal functions. In some cases, robotics has made it possible to manipulate variables in experiments that previously could only be investigated in correlational studies. Scientists studying the mating behavior of bowerbirds, for example, observed the females of the species repeatedly crouching during the attraction phase of mating (Patricelli et al., 2002). To examine the crouching variable in an experiment, they built a robotic version of a female bowerbird whose crouching actions could be remotely controlled. By systematically exposing male bowerbirds to crouching and non crouching behavior, and by varying the amount and frecuency of crouching, they where able to learn that males use females' crouching behavior as cues to initiate displays of their colorful plumage.
Experiments involving robotics birds may seem very far removed from any kind of practical application. However, projects such as this one help scientists and engineers learn more about how to build and program robots to behave like their living counterparts. And the resulting technology has tremendous potential for a wide variety of applications. Some impressive achievements in robotics include the following:
- Robots help stroke patients toward recovery by assisting them in exercise movement.
- Robotic surgical assistants help surgeons make remarkable gains in precision for some difficult surgeries.
- Robotic filling stations will keep satellites aloft.
- Miniature robots will assist tomorrow's soldiers.
- Robots can perform many duties too dangerous for humans (cleaning up toxic spills, finding and destroying land mines, cleaning up nuclear waste sites, etc.).
Some of the applications of robotics and artificial intelligence are so remarkable that is easy to losse sight of how they come to be. Keep in mind that the real power lies not in the machines, but in their creator―human cognition.
The World of Psychology p. 246-7
Do you like this post? Comment, share.