next up previous
Next: About this document Up: AN UNREASONABLE BOOK Previous: A Summary of Polemical

What Worries about Computers are Warranted?

Grumbling about Weizenbaum's mistakes and moralizing is not enough. Genuine worries prompted the book, and many people share them. Here are the genuine concerns that I can identify and the opinions of one computer scientist about their resolution: What is the danger that the computer will lead to a false model of man? What is the danger that computers will be misused? Can human-level artificial intelligence be achieved? What, if any, motivational characteristics will it have? Would the achievement of artificial intelligence be good or bad for humanity?

  1. Does the computer model lead to a false model of man.

    Historically, the mechanistic model of the life and the world followed animistic models in accordance with which, priests and medicine men tried to correct malfunctions of the environment and man by inducing spirits to behave better. Replacing them by mechanistic models replaced shamanism by medicine. Roszak explicitly would like to bring these models back, because he finds them more "human", but he ignores the sad fact that they don't work, because the world isn't constructed that way. The pre-computer mechanistic models of the mind were, in my opinion, unsuccessful,but I think the psychologists pursuing computational models of mental processes may eventually develop a really beneficial psychiatry.

    Philosophical and moral thinking hasn't yet found a model of man that relates human beliefs and purposes to the physical world in a plausible way. Some of the unsuccessful attempts have been more mechanistic than others. Both mechanistic and non-mechanistic models have led to great harm when made the basis of political ideology, because they have allowed tortuous reasoning to justify actions that simple human intuition regards as immoral. In my opinion, the relation between beliefs, purposes and wants to the physical world is a complicated but ultimately solvable problem. Computer models can help solve it, and can provide criteria that will enable us to reject false solutions. The latter is more important for now, and computer models are already hastening the decay of dialectical materialism in the Soviet Union.

  2. What is the danger that computers will be misused?

    Up to now, computers have been just another labor-saving technology. I don't agree with Weizenbaum's acceptance of the claim that our society would have been inundated by paper work without computers. Without computers, people would work a little harder and get a little less for their work. However, when home terminals become available, social changes of the magnitude of those produced by the telephone and automobile will occur. I have discussed them elsewhere, and I think they will be good - as were the changes produced by the automobile and the telephone. Tyranny comes from control of the police coupled with a tyrannical ideology; data banks will be a minor convenience. No dictatorship yet has been overthrown for lack of a data bank.

    One's estimate of whether technology will work out well in the future is correlated with one's view of how it worked out in the past. I think it has worked out well - e.g. cars were not a mistake - and am optimistic about the future. I feel that much current ideology is a combination of older anti-scientific and anti-technological views with new developments in the political technology of instigating and manipulating fears and guilt feelings.

  3. What motivations will artificial intelligence have?

    It will have what motivations we choose to give it. Those who finally create it should start by motivating it only to answer questions and should have the sense to ask for full pictures of the consequences of alternate actions rather than simply how to achieve a fixed goal, ignoring possible side-effects. Giving it human motivational structure with its shifting goals sensitive to physical state would require a deliberate effort beyond that required to make it behave intelligently.

  4. Will artificial intelligence be good or bad?

    Here we are talking about machines with the same range of intellectual abilities as are posessed by humans. However, the science fiction vision of robots with almost precisely the ability of a human is quite unlikely, because the next generation of computers or even hooking computers together would produce an intelligence that might be qualitatively like that of a human, but thousands of times faster. What would it be like to be able to put a hundred years thought into every decision? I think it is impossible to say whether qualitatively better answers would be obtained; we will have to try it and see.

    The achievement of above-human-level artificial intelligence will open to humanity an incredible variety of options. We cannot now fully envisage what these options will be, but it seems apparent that one of the first uses of high-level artificial intelligence will be to determine the consequences of alternate policies governing its use. I think the most likely variant is that man will use artificial intelligence to transform himself, but once its properties and the conequences of its use are known, we may decide not to use it. Science would then be a sport like mountain climbing; the point would be to discover the facts about the world using some stylized limited means. I wouldn't like that, but once man is confronted by the actuality of full AI, they may find our opinion as relevant to them as we would find the opinion of Pithecanthropus about whether subsequent evolution took the right course.

  5. What shouldn't computers be programmed to do.

    Obviously one shouldn't program computers to do things that shouldn't be done. Moreover, we shouldn't use programs to mislead ourselves or other people. Apart from that, I find none of Weizenbaum's examples convincing. However, I doubt the advisability of making robots with human-like motivational and emotional structures that might have rights and duties independently of humans. Moreover, I think it might be dangerous to make a machine that evolved intelligence by responding to a program of rewards and punishments unless its trainers understand the intellectual and motivational structure being evolved.

    All these questions merit and have received more extensive discussion, but I think the only rational policy now is to expect the people confronted by the problem to understand their best interests better than we now can. Even if full AI were to arrive next year, this would be right. Correct decisions will require an intense effort that cannot be mobilized to consider an eventuality that is still remote. Imagine asking the presidential candidates to debate on TV what each of them would do about each of the forms that full AI might take.


McCulloch,W.S.(1956) ``Toward some circuitry of ethical robots or an observational science of the genesis of social evaluation in the mind-like behavior of artifacts.'' Acta Biotheoretica,XI,parts 3/4, 147-156

Weizenbaum, Joseph (1965) ``ELIZA--a computer program for the study of natural language communication between man and machine'', Communications of the Association for Computing Machinery,9, No. 1, 36-45.

John McCarthy
Computer Science Department
Stanford, California 94305

next up previous
Next: About this document Up: AN UNREASONABLE BOOK Previous: A Summary of Polemical

John McCarthy
Tue Oct 17 20:28:09 PDT 2000