Saturday, August 31, 2019
More AI point-missing
A pretty good article on AI by Gary Smith, but it still doesn't quite hit the biggest point.
Consider the challenges identified by Stanford computer science professor Terry Winograd, which have come to be known as Winograd schemas. For example, what does the word “it” refer to in this sentence? I can’t cut that tree down with that axe; it is too [thick/small]. If the bracketed word is “thick,” then it refers to the tree; if the bracketed word is “small,” then it refers to the axe. Sentences like these are understood immediately by humans but are very difficult for computers because they do not have the real-world experience to place words in context.The overall point about context is excellent, but the specific example doesn't work well. In fact a person who has actually dealt with trees would have the same problem as the computer, for different reasons. A "thick" axe isn't a good wedge. A "small" tree can be hard to cut because it simply bends over instead of holding still. In this case the pronoun is just incurably ambiguous. The second phrase must specify tree or ax. Winograd was making assumptions based on lack of experience. Smith hits the problem of random clusters accurately and clearly. (I've covered it graphically.) But again the problem isn't the computer's lack of judgment. It's the lack of real-world experience by the people who trust the computer to make their decisions. They don't recognize that the computer is leading them astray. A manager who had started his career as a salesman in a real store wouldn't believe the computer's decisions about marketing, and would correctly trust his own judgment. If we had more people with hands-on experience in positions of power, they wouldn't misuse the computer. We have too many managers who don't know more than the computer. The solution would be more hands-on education.... but we're not going to get the solution because the people with REAL power know exactly what they're doing. They want to prevent most people from gaining experience and judgment. They know that experience enables people to spot verbal lies.
Labels: AI point-missing, Carver, Experiential education
¶ 12:14 AM
