4.8 ESTRATEGIAS DE SERVICIOS
4.8.3 Estrategia de precios
By the mid-1960s AI was well prepared for further advances. Flushed with early successes it was poised to make rapid progress during the rest of the 1960s and 1970s. Indeed, many people made enthusiastic predictions. For example, in a 1957 talk16 Herb Simon predicted that within ten years “a digital computer will be the world’s chess champion unless the rules bar it from competition.” He made three other predictions too. Within ten years computers would compose music, prove a mathematical theorem, and embody a psychological theory as a program. He said “it is not my aim to surprise or shock you. . . but the simplest way I can summarize is to say that there are now in the world machines that think, that learn and that create. Moreover, their ability to do these things is going to increase rapidly until – in a visible future – the range of problems they can handle will be coextensive with the range to which the human mind has been applied.”17 Later Simon said that his predictions were part of an attempt “to give some feeling for what computers would mean” to society.
One could argue that Simon’s predictions about computers composing music and proving a mathematical theorem were realized soon after he made them, but a computer chess champion was not to emerge until forty years later. And, we are still far, I think, from achieving things “coextensive” with what the human mind can achieve.
Simon was not alone in being optimistic. According to Hubert Dreyfus, “Marvin Minsky, head of MITs Artificial Intelligence Laboratory, declared in a 1968 press release for Stanley Kubrick’s movie,2001: A Space Odyssey, that ‘in 30 years we should have machines whose intelligence is comparable to man’s.’ ”18 The difficulty in assessing these sorts of predictions is that “human-level intelligence” is multifaceted. By the year 2000, AI programs did
outperform humans in many intellectual feats while still having a long way to go in most others.
Even so, what had already been accomplished was an impressive start. More important perhaps than the specific demonstrations of intelligent behavior by machines was the technical base developed during the 1950s and early 1960s. AI researchers now had the means to represent knowledge by encoding it in networks, as logical formulas, or in other symbol structures tailored to specific problem areas. Furthermore, they had accumulated
experience with heuristic search and other techniques for manipulating and using that knowledge. Also, researchers now had new programming languages,
IPL,LISP, andPOP-2, that made it easier to write symbol-processing programs. Complementing all of this symbol-processing technology were neural networks and related statistical approaches to pattern recognition. These technical assets, along with the organizational and financial ones, provided a solid base for the next stage of AI’s development.
Notes
1.For McCarthy’s own history of the development ofLISP, see
http://www-formal.stanford.edu/jmc/history/lisp.html. Also see Herbert Stoyan’s history of
LISPathttp://www8.informatik.uni-erlangen.de/html/lisp-enter.html. [156]
2.See McCarthy’s memo proposing how to build a time-sharing system at
http://www-formal.stanford.edu/jmc/history/timesharing-memo.html. [157]
3.For more about these early days of computing at MIT and of time-sharing work there (among other things), see the interview with John McCarthy conducted by William Aspray of the Charles Babbage Institute on March 2, 1989. It is available online at
http://www.cbi.umn.edu/oh/display.phtml?id=92. [157]
4.From an interview conducted by Arthur L. Norberg on November 1, 1989, for the Charles Babbage Institute. Available online at
http://www.cbi.umn.edu/oh/display.phtml?id=107. [157]
5.For a history of AI work in the lab up to 1973, see Lester Earnest (ed.), “Final Report: The First Ten Years of Artificial Intelligence Research at Stanford,” Stanford Artificial Intelligence Laboratory Memo AIM-228 and Stanford Computer Science Department Report No. STAN-CS-74-409, July 1973. (Available online at
http://www-db.stanford.edu/pub/cstr/reports/cs/tr/74/409/CS-TR-74-409.pdf.) For other SAIL history, see “SAIL Away” by Les Earnest at
http://www.stanford.edu/∼learnest/sailaway.htm. [158]
6.A textscript of the interview can be found online at
http://www.aiai.ed.ac.uk/events/ccs2002/CCS-early-british-ai-dmichie.pdf. [158]
7.Donald Michie, “Experiments on the Mechanisation of Game Learning: 1.
Characterization of the Model and its Parameters,”Computer Journal, Vol. 1, pp. 232–263, 1963. [158]
8.For a history of these Edinburgh groups, see Jim Howe’s online 1994 article “Artificial Intelligence at Edinburgh University: A Perspective” at
http://www.dai.ed.ac.uk/AI at Edinburgh perspective.html. [160]
9.National Research Council,Funding a Revolution: Government Support for Computing Research, Washington, DC: National Academy Press, 1999. (An html version of this book, which contains a rather conservative account of AI history, is available from
http://www.nap.edu/catalog.php?record id=6323#toc.) [160]
10.From “An Interview with John McCarthy,” conducted by William Aspray on 2 March 1989, Palo Alto, CA, Charles Babbage Institute, The Center for the History of Information Processing, University of Minnesota, Minneapolis. [160]
11.J. C. R. Licklider, “Man–Computer Symbiosis,”IRE Transactions on Human Factors in Electronics, HFE-1, pp. 4–11, 1960. Available online athttp://memex.org/licklider.html. [161]
12.J. C. R. Licklider, “The Early Years: Founding IPTO,” p. 220 in Thomas C. Bartee (ed.),Expert Systems and Artificial Intelligence: Applications And Management, Indianapolis: Howard W. Sams, 1988. [161]
13.Paul Edwards,The Closed World: Computers and the Politics of Discourse in Cold War America, p. 270, Cambridge, MA: MIT Press, 1996. [161]
14.From “An Interview with John McCarthy,”op. cit. [162]
15.Lester Earnest (ed.), “Final Report: The First Ten Years of Artificial Intelligence Research at Stanford,” Stanford Artificial Intelligence Laboratory Memo AIM-228 and Stanford Computer Science Department Report No. STAN-CS-74-409, July 1973. (Available online athttp://www-db.stanford.edu/pub/cstr/reports/cs/tr/74/409/CS-TR-74-409.pdf.) [162]
16.12th National Meeting of the Operations Research Society (ORSA) in Pittsburgh. [163]
17.The published version of this talk is in Herbert Simon and Allen Newell, “Heuristic Problem Solving: The Next Advance in Operations Research,”Operations Research, Vol. 6, January–February 1958. [163]
18.Hubert L. Dreyfus, “Overcoming the Myth of the Mental,”Topoi, Vol. 25, pp. 43–49, 2006. [163]
Part III
Efflorescence: Mid-1960s to
Mid-1970s
During the 1960s and well into the 1970s, AI research blossomed and progress seemed rapid. The laboratories established at MIT, Carnegie Mellon, Stanford, SRI, and Edinburgh expanded, and several new groups got started at other universities and companies. Achievements during the preceding years, even though modest in retrospect, were exciting and full of promise, which enticed several new people into the field, myself included. Many of us were just as optimistic about success as Herb Simon and Marvin Minsky were when they made their predictions about rapid progress.
AI entered a period of flowering that led to many new and important inventions. Several ideas originated in the context of Ph.D. dissertation research projects. Others emerged from research laboratories and from individual investigators wrestling with theoretical problems. In this part, I’ll highlight some of the important projects and research results. Although not a complete account, they typify much of what was going on in AI during the period. I’ll begin with work in computer vision.
Chapter 9
Computer Vision
Sighted humans get much of their information through vision. That part of AI called “computer vision” (or, sometimes, “machine vision”) deals with giving computers this ability. Most computer vision work is based on processing two-dimensional images gathered from a three-dimensional world – images gathered by one or more television cameras, for example. Because the images are two-dimensional projections of a three-dimensional scene, the imaging process loses information. That is, different three-dimensional scenes might produce the same two-dimensional image. Thus, the problem of reconstructing the scene faithfully from an image is impossible in principle.
Yet, people and other animals manage very well in a three-dimensional world. They seem to be able to interpret the two-dimensional images formed on their retinas in a way that gives them reasonably accurate and useful information about their environments.
Stereo vision, using two eyes, helps provide depth information. Computer vision too can use “stereopsis” by employing two or more differently located cameras looking at the same scene. (The same effect can be achieved by having one camera move to different positions.) When two cameras are used, for example, the images formed by them are slightly displaced with respect to each other, and this displacement can be used to calculate distances to various parts of the scene. The computation involves comparing the relative locations in the images that correspond to the objects in the scene for which depth measurements are desired. This “correspondence problem” has been solved in various ways, one of which is to seek high correlations between small areas in one image with small areas in the other. Once the “disparity” of the location of an image feature in the two images is known, the distance to that part of the scene giving rise to this image feature can be calculated by using trigonometric calculations (which I won’t go into here.)1
other cues besides stereo vision. Some of these cues are inherent in a single image, and I’ll be describing these in later chapters. Even more importantly, background knowledge about the kinds of objects one is likely to see accounts for much of our ability to interpret images. Consider the image shown in Fig. 9.1for example.
Figure 9.1: Two tables. (Illustration courtesy of Michael Bach.) Most people would describe this image as being of two tables, one long and narrow and the other more-or-less square. Yet, if you measure the actual table tops in the image itself, you might be surprised to find that they are exactly the same size and shape! (The illustration is based on an illusion called “turning the tables” by the psychologist Roger Shepherd and is adapted from Michael Bach’s version of Shepherd’s diagram. If you visit Bach’s Web site,http://www.michaelbach.de/ot/sze shepardTables/, you can watch while one table top moves over to the other without changing shape.)
Something apart from the image provides us with information that induces us to make inferences about the shapes of the three-dimensional tables captured in the two-dimensional image shown in Fig. 9.1. As we shall see, that extra information consists of two things: knowledge about the image-forming process under various lighting conditions and knowledge about the kinds of things and their surfaces that occur in our three-dimensional world. If we could endow computers with this sort of knowledge, perhaps they too would be able to see.
9.1
Hints from Biology
There has been a steady flow of information back and forth between scientists attempting to understand how vision works in animals and engineers working on computer vision. An early example of work at the intersection of these two interests was described in an article titled “What the Frog’s Eye Tells the Frog’s Brain”2 by four scientists at MIT. Guided by previous biological work, the four, Jerome Lettvin, H. R. Maturana, Warren McCulloch, and Walter Pitts, probed the parts of the frog’s brain that processed images. They found that the frog’s visual system consisted of “detectors” that responded only to certain kinds of things in its visual field. It had detectors for small, moving convex objects (such as flies) and for a sudden darkening of illumination (such as might be caused by a looming predator). These, together with a couple of other simple detectors, gave the frog information about food and danger. In particular, the frog’s visual system did not, apparently, construct a complete three-dimensional model of its visual scene. As the authors wrote,
The frog does not seem to see or, at any rate, is not concerned with the detail of stationary parts of the world around him. He will starve to death surrounded by food if it is not moving. His choice of food is determined only by size and movement. He will leap to capture any object the size of an insect or worm, providing it moves like one. He can be fooled easily not only by a bit of dangled meat but by any moving small object. His sex life is conducted by sound and touch. His choice of paths in escaping enemies does not seem to be governed by anything more devious than leaping to where it is darker. Since he is equally at home in water and on land, why should it matter where he lights after jumping or what particular direction he takes?
Other experiments produced further information about how the brain processes visual images. Neurophysiologists David Hubel (1926– ) and Torsten Wiesel (1924– ) performed a series of experiments, beginning around 1958, which showed that certain neurons in the mammalian visual cortex responded selectively to images and parts of images of specific shapes. In 1959 they implanted microelectrodes in the primary visual cortex of an anesthetized cat. They found that certain neurons fired rapidly when the cat was shown images of small lines at one angle and that other neurons fired rapidly in response to small lines at another angle. In fact, they could make a “map” of this area of the cat’s brain, relating neuron location to line angle. They called these neurons “simple cells” – to be distinguished from other cells, called “complex cells,” that responded selectively to lines moving in a certain direction. Later work revealed that other neurons were specialized to respond to images containing more complex shapes such as corners, longer lines, and large edges.3 They found that similar specialized neurons also existed in the brains
of monkeys.4 Hubel and Wiesel were awarded the Nobel Prize in Physiology or Medicine in 1981 (jointly with Roger Sperry for other work).5
As I’ll describe in later sections, computer vision researchers were developing methods for extracting lines (both large and small) from images. Hubel and Wiesel’s work helped to confirm their view that finding lines in images was an important part of the visual process. Yet, straight lines seldom occur in the natural environments in which cats (and humans) evolved, so why do they (and we) have neurons specialized for detecting them? In fact, in 1992 the neuroscientists Horace B. Barlow and David J. Tolhurst wrote a paper titled “Why Do You Have Edge Detectors?”6As a possible answer to this question, Anthony J. Bell and Terrence J. Sejnowski later showed
mathematically that natural scenes can be analyzed as a weighted summation of small edges even though the scenes themselves do not have obvious edges.7