Comments
Description
Transcript
Problem Solving by Computer
262 Chapter 7 Thought, Language, and Intelligence under way. He was about to alert his superiors to launch a counterattack on the United States when it occurred to him that if this were a real nuclear attack, it would involve many more than five missiles. Fortunately for everyone, he realized that the “attack” was a false alarm (Hoffman, 1999). As this near-disaster shows, the absence of symptoms or events can sometimes provide important evidence for or against a hypothesis. Compared with evidence that is present, however, symptoms or events that do not occur are less likely to be noticed (Hunt & Rouse, 1981). People have a difficult time using the absence of evidence to help eliminate hypotheses from consideration (Hyman, 2002). In the “trapped kitten” case, when the “meowing” stopped for several days after the stove was unplugged and reconnected, rescuers assumed that the animal was frightened into silence. They ignored the possibility that their hypothesis was incorrect in the first place. FIGURE 7.8 Two Creative Solutions to the Nine-Dot Problem Many people find problems like this difficult because mental sets create artificial limits on the range of solutions. In this case, the mental sets involve the tendency to draw within the frame of the dots and to draw through the middle of each dot. As shown here, however, there are other possibilities. Confirmation Bias Anyone who has had a series of medical tests knows that diagnosis is not a one-shot affair. Instead, physicians choose their first hypothesis on the basis of observed symptoms and then order tests or evaluate additional symptoms to confirm or eliminate that hypothesis (Trillin, 2001). This process can be distorted by a confirmation bias: Humans have a strong bias to confirm rather than reject the hypothesis they have chosen, even in the face of strong evidence against the hypothesis. In other words, people are quite willing to perceive and accept data that support their hypothesis, but they tend to ignore information that is inconsistent with it (Groopman, 2000). Confirmation bias may be seen as a form of the anchoring heuristic. Once you’ve “anchored” to your first hypothesis, you may be unwilling to abandon it. The would-be rescuers of John Gatiss’s “trapped kitten” were so intent on their efforts to pinpoint its location that they never stopped to question its existence. Similarly, as described in the chapter on social psychology, we tend to look for and pay extra attention to information that is consistent with our first impressions of other people. This tendency can create positive or negative bias in, say, a teacher’s views of children’s cognitive abilities or an interviewer’s judgments of a job candidate’s skills (Jussim & Eccles, 1992; Reich, 2004). (For a summary of problem solving and its pitfalls, see “In Review: Solving Problems.”) Problem Solving by Computer Researchers have created artificial limbs, retinas, cochleas, and even hearts to help disabled people move, see, hear, and live more normally. They are developing artificial brains, too, in the form of computer systems that not only see, hear, and manipulate objects but also reason and solve problems. These systems are the product of research in artificial intelligence (AI), a field that seeks to develop computers that imitate the processes of human perception and thought. confirmation bias The tendency to pay more attention to evidence in support of one’s hypothesis about a problem than to evidence that refutes that hypothesis. An IBM computer known as Deep Blue has won chess games against the world’s best chess masters. This result is not surprising, because chess is a clearly defined, logical game at which computers can perform effectively. However, it is precisely their reliance on logic and formulas that accounts for the shortcomings of today’s artificial intelligence systems. For example, these systems are successful only in narrowly defined fields, not in general problem solving. This limitation stems from the fact that AI systems are based on logical symbolic manipulations that depend on “if-then” rules. Unfortunately, it is difficult to tell a computer how to recognize the “if” condition in the real world. Consider this simple “if-then” rule: “If it is a clock, then set it.” Humans recognize all kinds of clocks because they have the natural concept of “clock,” but computers are still not very good at forming natural concepts. Doing so requires putting into the same category many examples that have very different physical features, from your bedside digital alarm clock to London’s Big Ben. artificial intelligence (AI) The field that studies how to program computers to imitate the products of human perception, understanding, and thought. Neural Network Models Recognizing the problems posed by the need to teach computers to form natural concepts, many researchers in AI have moved toward a connectionist, or neural network, approach. This approach uses computers to simulate Symbolic Reasoning and Computer Logic 263 in review Problem Solving SOLVING PROBLEMS Steps Pitfalls Remedies Define the problem. Inexperience: the tendency to see each problem as unique Gain experience and practice in seeing the similarity between present problems and previous problems. Form hypotheses about solutions. Availability heuristic: the tendency to recall the hypothesis or solution that is most available to memory Anchoring heuristic, or mental set: the tendency to anchor on the first solution or hypothesis, and not adjust your beliefs in light of new evidence or failures of the current approach Force yourself to write down, and carefully consider, many different hypotheses. Break the mental set, stop, and try a fresh approach. Test hypotheses. The tendency to ignore negative evidence In evaluating a hypothesis, consider the things you should see (but don’t) if the hypothesis were true. Look for disconfirming evidence that, if found, would show your hypothesis to be false. Confirmation bias: the tendency to seek only evidence that confirms your hypothesis ? 1. People stranded without water could use their shoes to collect rain, but they may not do so because of an obstacle to problem solving called . 2. Because of the heuristic, once sellers set a value on their house, they may refuse to take much less for it. 3. If you tackle a massive problem one small step at a time, you are using an approach called . the information processing taking place at many different, but interconnected, locations in the brain. Neural network models have helped researchers develop computers that are able to recognize voices, understand speech, read print, guide missiles, and perform many other complex tasks (Ashcraft, 2006). Some of these computer simulations are being used to improve speech recognition software and to test theories of how infants learn to recognize speech (e.g., Roy & Pentland, 2002; Sroka & Braida, 2005). Others have been used to improve on human decision making. One program, called PAPNET, can outperform human technicians at detecting abnormal cells in smears collected during cervical examinations (Kok & Boon, 1996). Indeed, computerized expert systems can now perform as well as humans, and sometimes better, at solving complex problems in medical diagnosis and business decision making (Khan et al., 2001; Workman, 2004). Unfortunately, however, most computer models of neural networks still fall well short of the capacities of the human perceptual system. For example, computers are slow to learn how to classify visual patterns, which has led to disappointment in efforts to develop computerized face recognition systems capable of identifying terrorists and other criminals in public places (Feder, 2004). But even though neural networks are far from perfect “thinking machines,” they are sure to play an important role in