1.6. ARTIFICIAL INTELLIGENCEThe term Artificial Intelligence refers to work that tries to make computers replicated various aspects of human intelligence, such as understanding the meaning of pictures, understanding the meaning of speech, translating text from one human language into another, playing games such as chess and checkers, and so forth. Another term used for such efforts is Machine Intelligence and, personally, I prefer it for being more precise. It also avoids the pun "Natural Stupidity always beats Artificial Intelligence." But the term Artificial Intelligence (AI) is too well entrenched in the public mind and I will stay with it. In the early days of computing there was great optimism for solving the AI problems. Sixty years later there is a more sobering view. What has happened is that we have seen enormous progress in solving specific problems using methods matched to the problem. This is usually called Narrow Artificial Intelligence (Narrow AI). However, no progress has been made in replicating human intelligence in a general form. This field is usually called General Artificial Intelligence (General AI) and it is certainly more glamorous than Narrow AI. It also has the appeal of "getting something for nothing", solving problems without having really to understand them. As a result, there are always claims that "General AI is around the corner". But after hearing such claims for half a century I am not going to hold my breadth and, instead, I will focus on the successes of Narrow AI. One very "hot" area of Narrow AI is finding what people like. Merchants would love to make you offers that you are likely to accept and for that they have to know your tastes. A lot of progress has been made (and a lot remains to be done) by using statistics to analyze past behavior in order to try to predict future behavior. We discuss such issues in Chapter 2, Finding what People Like. Another area that had captured the public attention is computers that play game. When IBM's Deep Blue beat the human chess world champion in 1997 there was a lot of media coverage and a lot of articles on how computers are now smarter than humans. The latter claims were way off the mark. Deep Blue played using a naive "brute force" approach and as a result of its speed could match the human level of chess playing. All computer game playing programs rely on variations of this approach. We discuss several of them in Chapter 3, Games Computers Play. One of the earliest efforts in making computers that replicate a human behavior dealt with reading by computers. Turing was interested in the subject in the 1940's and it continues to be an active area of research today. The machine starts with the image of a block of text and then it tries to identify the letters there and eventually convert the image into a string of characters. This is the way that Google includes old books in their databases. We discuss the topic in Chapter 4, Teaching Computers to Read. In that chapter we also cover barcodes a technology that relies on a "computer-friendly" alphabet. As a result barcode scanners are much cheaper than text scanners. Anyone who has ever called a business phone number has come across synthetic speech and in the last decade also to speech recognition: "If want to hear this message in English please press or say one". The use of automation in telephone services is so pervasive that there is a web site that provides lists of numbers answered by humans (http://gethuman.com/). Speech synthesis and voice recognition are the subjects of Chapter 5, Teaching Computers to Talk and to Listen. It is worth pointing out that speech synthesis turned out to be a much easier problem than speech understanding. Making computers understand what they see (Image Analysis or Image Understanding) has been one of the bigger challenges in Artificial Intelligence. We touched briefly upon this issue in Section 1.5 and we discuss it at length in Chapter 6, Making Sense of what a Computer Sees. Applications include surveillance because the computer never gets bored in the long periods when nothing happens. A system for detecting forest fires based on video input has been deployed successfully in Turkey and other Mediterranean countries. On the other hand face recognition has not lived up to the expectation. We will explain why. Interestingly, composing pictures by computer (Computer Graphics) turned out to be much easier than analyzing pictures, a parallel to the speech synthesis versus speech understanding. Computers have been very good in constructing pictures out of numbers, not only in creating the displays we see in modern animation but also in medical applications, such as CAT scans (CAT stands for Cross Axial Tomography). We discuss Synthetic Images in Chapter 7. Robots are computer controlled machines and can be either entirely autonomous or controlled remotely by human operators. We will provide examples from both industriral and home use, and also discuss cars driven only by robots (the DARPA challenge). Robotics will be the subject of Chapter 8. Finally, in Chapter 9, Predictions, we deal with the ability of computers to make predictions, from the weather to the stock market. It turns out that large complex systems (such as the weather) are inherently unpredictable and they are called Chaotic. This feature has a lot of implications, both practical and philosophical.
|