Last updated: 4/23/03
Link back to course Welcome

Artificial Intelligence
(additional issue)

1. What is artificial intelligence? Of course, there are lots of jokes questioning whether humans are intelligent in the first place. But humans are far better at everyday reasoning than computer are. On the other hand, computers are getting better. The human brain has 10 billion nerves. Of course we don't know whether a  nerve is the equivalent of a bit, or a byte, or something else, but it does not seem that storage capacity could account for the differences between humans and computers all by itself. Here are some of the hypotheses (guesses) as to why humans are so much better at reasoning:
   1. Even though our "hardware" in the brain, sometimes called "wetware," is much slower, it is different than the Von Neumann architecture like that in the Paper and Pencil computer, and our neurons may be much more efficient at everyday reasoning. Research is going on into the nature of neurons, and people are attempting to build computer systems that act more like the human brain. One example of such systems is the "neural network," a network of interconnected software modules that mimic some aspects of neurons, and can display some characteristics of learning from experience.
   2. Another possibility is that, even if our hardware isn't better, we do many more things at the same time ("multitasking" or "parallel computing") and so out-compute computers this way. Each of our brain's neurons is a processor. Again, computer systems are being constructed that do many more things in parallel.
   3. A third possibility is that, even by the time we are five or six years old, we have an enormous store of background information that helps us make sense of new information, while a computer has to learn everything from ground zero every time it starts a new program. There are attempts to program massive amounts of background information into computers.
   4. The fourth possibility that we will mention here is that the structure of our minds may be more flexible. Basically, we have an unconscious that takes care of many things without our attention, and which may even do some thinking for us, for example during dreams, or before a flash of insight (an aha! experience), and a conscious mind which is self-aware and can direct itself to focus on a situation. It is not clear how to mimic this in a computer, and in fact it is not at all clear today how human consciousness works.
2. Part of the problem with understanding what artificial intelligence means is that we only have a very general idea of what intelligence itself means. Generally, the most basic definition is an ability to use brain power to adapt to changes in the total environment. By "total environment" we mean not only the biological or living environment, as we normally use the word, but also the effects of our civilization, such as buildings and cars, and the effects of other people (the social and personal environment) but even our own internal environment, our own thoughts and actions. We do have a measure of intelligence, the IQ, and an IQ test, but this is much criticized as not measuring the ability to adapt, but instead the results of having adapted in the past, and to a specific environment, generally a Western white middle-class environment. So aside from this specific meaning (the score on a particular test that may or may not measure what we want it to), we mean the ability to function well in the face of change and incomplete information, to be able to integrate information from a wide variety of sources, to be able to sort through information acquired in the past and to focus relevant facts from the past on the current situation, to learn, and so forth. Are all of these facets related to each other, or are they all independent, and is any one list complete? We don't know. I believe that we will learn all of this and more in the future, but for now we don't know exactly what intelligence itself is.
3. Historically, we are becoming more demanding before we call something "intelligent." For example, before computing we thought it required intelligence to add. But now that we have $2 calculators that can add, we don't think this is intelligent any more. Similarly, proofreading and outlining functions were originally thought to be intelligent. And now that we have computer systems that act as if they are learning, at least in a simple way, simple learning is not thought to be intelligent. Clearly, we are much more able to deal with conflicting and incomplete information than computers are, and we are much better at learning.
4. Some commentators argue that computer can never be intelligent, that this is beyond what any device following a program (a list of instructions) can ever be intelligent in the way that humans are. As a skeptical scientist, I have seen many such arguments disproven. The only real proof in this area would be a computer that actually is intelligent; by this argument, if we do not have an intelligent computer yet, that is just because we haven't learned the right way yet. Even supposing that we never have an intelligent computer, computers can still be very helpful, by doing for us the things that we are not good at, such as following directions exactly.
5. How would we tell if a computer is intelligent? The usual test is "the Turing test," originated by the English computer scientist Alan Turing. In this test, a computer and a human are put behind a wall, and a human or panel of humans ask questions, which are answered by means of printing on paper or some other means that can be done by both humans and computers. First a question to the human, then one to the computer, etc. This seems like a good test, and is run regularly. It is depressing that even simple computer programs can fool human beings. A famous example is the "Eliza" program - see page INT-10 in Computers, Technology and Society.
6. The "agent interface" is an example of trying to apply simple intelligence to the interactions between humans and computers. The "Office Assistant" is an example of an agent interface, although since I don't like the Office Assistant (the paper clip guy), I hope that this is an early and crude example.
7. There is no doubt that even moderate artificial intelligence within a computer would have a major economic and social impact. Computers would be much better at working with humans, and programs with artificial intelligence interfaces would quickly take over the market. Humans would no longer be employable unless they used even higher-level skills, just as machines took over from human muscle, and computers from much of accounting and proofreading. So far, I think that technology, while it has created problems no doubt, has also given us much more freedom of choice than we have had before, and it has also allowed our population to soar to the point where we are a massive global threat to other species. Would artificial intelligence be like this, where we would more or less voluntarily adopt it, or would the competition between humans and computers lead to our hating and rejecting computers? Fascinating? Stay tuned and just maybe we'll find out!
8. Currently, computers have traits that are very much different from those of humans: they to follow directions mindlessly and tirelessly, and their memories do not change over time (unless the fail completely). Presently, human and computer abilities are complementary, and computers can be a significant help to humans.