Chapter Nine
Do You Think Artificial Intelligence Will Equal Or Surpass Human Intelligence?
Adams (U.C. Los Angeles): I think the first thing I would say is that I do not have a very informed opinion on artificial intelligence. I assume that in some respect artificial intelligence has already surpassed human intelligence. My computer can do some things that I can't. But do I think that someday computers will be able to do every task that humans can handle? No, I don't really think that. But that may be as much of an expression of prejudice as anything else. It's not based on a particular knowledge of what computers can do. That's a question really outside my area of expertise. Arntzenius (University of Southern California): It depends a bit on what you mean by intelligence. If you mean the ability to do IQ tests well then I certainly think that we will be able to design computers that will do the tests better than we can. Why? Well we've been able to program computers to do arithmetic better than we do; they certainly play chess better than I do. In practice we seem to be a very complicated and very efficiently designed machine. I doubt very much if our explicit design will be able to construct something that in almost all areas outperforms us. I'm not even convinced that the hardest problem is to design something that has the mental capacity that we do. Beckman (Harvey Mudd): There are several different issues to be considered in artificial intelligence. As you've phrased your question in terms of "intelligence," we are forced to ask how intelligence should be assessed. If, for example, we accept the idea that intelligence should be measured by how rapidly something can perform complex mathematical tasks or store and retrieve mathematical information, then computers have already beaten the human mind by a long shot. However, if we interpret intelligence in some more complex way--say, translating between human natural languages or making design decisions based on more than technical factors--it's not entirely clear whether computers will ever beat the human mind. If we ask the question in terms of "consciousness" rather than "intelligence," then I don't hold out much hope that computers will ever replicate human consciousness; in other words, I think that artificial intelligence will always be "artificial" in significant ways. Churchland, Paul (U.C. San Diego): I think that it will surpass human intelligence. I think that in some dimensions it surpassed human intelligence twenty years ago. However, it surpassed it in only a very narrow capacity--the capacity for a sheer repetitive computation like doing long division, or multiplication, or addition or things like that but intelligence is a very much broader capacity than that. I think it will take fifty or a hundred years before we understand the human brain fully. When we understand how the human brain works, I think it will then be relatively straightforward, though it will be difficult. It will be conceptually a straightforward technological matter to make an artificial system which can do the things that we do. I don't think we will do that however. It's too easy to make human intelligence already. You only need a loving couple to do it. So we're not going to put up millions of dollars to make artificial humans. What we will do instead is to take arti icial intelligence systems for some scientific purpose that we create some subset of the human capacities or perhaps will show some cognitive feature that we don't have at all. After all there are many more kinds of brains possible than just the human brain. I fully expect in the fairly near future for artificial intelligence to exceed the humans in many dimensions. I don't know how this is all going to come out. I think it's going to be a very exciting and interesting adventure and I'm not entirely comfortable with every aspect of it, but I think it will happen. Cohon (Stanford University): I can't predict what kinds of machines will be built in the future. On the one hand, I am inclined to think that, since the human brain is made of matter and it can think, it is possible to make other things out of matter that do the same things. On the other hand, much of what we classify as intelligent is socially defined and can only occur within a social context; this is especially true of speech. Consequently, it may be that no real machine intelligence is possible in the absence of some sort of machine community or society of machines. Anyway, while scholars are talking about science fiction scenarios, real researchers in machine intelligence are very far from understanding what human intelligence is, so at present it is impossible to predict whether machine intelligence can be made to equal it. For example, it is not understood how human beings recognize faces or understand speech of unknown persons, and psychologists are only beginning to figure out how people make sens of information that is presented to them in written form. Until we know what human intelligence is it is impossible to say whether machines will be able to duplicate it or surpass it. Copp (U.C. Davis): In some respects, yes; in others no, I doubt it will surpass human creativity. But I am only projecting up to 50 years. Beyond that, who knows? Dreyfus (U.C. Berkeley): I have written two books on this subject. There are two kinds of artificial intelligence. The first kind, which started in about 1960, was devoted to using computers that were called physical symbol systems. The computer would have in it symbols that represented features of the world and the programs of the computer would be used to make inferences and deduce conclusions from this representation of features. I said in 1965, and in my book in 1972, that this kind of artificial intelligence would not work because of our way of being in the world is not having in our mind representations and features. It turns out that I think (and lots of other people now are beginning to think) that I was right, that it is failing. There was an article that had A.I. on the cover and quoted me and agreed with me that symbolic A.I. did not succeed. But now there is a new kind of A.I. using computers doing what is called simulated neural networks. I think that that will never produce full human intelli ence, but I think that it is not philosophically wrong like the symbolic A.I. but I do not think that it will work because the brain is too complicated and we do not know how it is wired up, so we can't make a simulated network that is enough like it even if we could. I think that the fact that we have bodies and move around in a world and have a culture is part of the way that our neural network gets tuned the way it is tuned and a computer that just had a neural network and passively took in what is called input vectors and paired them with output vectors [will not] have our kind of intelligence. Fischer (U.C. Riverside): In certain ways, such as calculations, computers are already equal, if not better. They are also continually progressing in mechanics. However, I remain skeptical as to whether computers will ever be as insightful or as creative as the human mind. Friedman (U.C. Davis): No, never! Because we lead from it, and so we will be that much smarter. I've believed in manmachine relations for a long time. Griesemer (U.C. Davis): I'm not convinced that artificial intelligence is intelligence, so I don't think there's yet a question about whether it will bypass human intelligence. I think intelligence is a property of certain biological entities, so whatever computers are capable of, it isn't intelligence (unless computers are capable of being certain sorts of biological entities!). It's merely by analogy that we call what computing machines do thinking. Jolley (U.C. San Diego): I am not well versed in this debate, but no I don't think that artificial intelligence will equal or bypass human intelligence except in very limited spheres (such as the ability to perform calculations at fantastic speed). My reasons are those which Descartes gives in the Discourses of Method, Part V. Jubien (U.C. Davis): If "artificial intelligence" just refers to the capacities of computers, then I think it already exceeds the capacities of human intelligence in certain ways (e.g., speed of computation). I don't think computers will ever have fully "human" intelligence because I don't think they will ever have mental experiences akin to those that humans have. Kalish (U.C. Los Angeles): There are things which computers can do now which human beings can't do and the speed with which you can do computations and things like that are incredible. Also computers can store an enormous amount of information in its memories and you can get it back. On the other hand, it is quite well known that there are problems which you cannot prove and that there is no algorithm which will ever solve them. Human ingenuity is the only way they will ever be solved. So these are two entities where the human mind and the artificial machine differ. They both have enormous qualities and it is not a matter of trying to say that there are two things in which one is a little better than the other. There are certain things that one can do that the other can't and we are getting better and better. My gosh the things that can be done now and the way you can communicate with people is fascinating. So let me put it this way, I don't think that any person of your generation who doesn't become comput r competent is going to be able to compete in this life. There are mathematical problems that only the human mind will be able to answer because we can prove that there is no way we can program a machine to answer the question. Kaplan (U.C. Los Angeles): There is the so called Turing test. When you carry on a conversation and you can't tell if it's a machine or a person. I don't really have a view as to whether we will be able to create machines that will pass the Turing test. It's clear that machines can already do tasks which require a kind of intelligence, much better than we can. I use a spell checker because it is a better speller than I am and quicker. I don't know of any machine that is as creative as I am. I am very skeptical to whether we will be able to do it, unless we start to build biological machines. Lambert (U.C. Irvine): Well, there are several things to think about it. First, it's a difficult thing to say what human intelligence is and it's hard to tell whether artificial intelligence will surpass human intelligence or not. We are not even clear what human intelligence is. But if it means, for example, that machines will be able to do certain things better than human beings would do intellectual things, well they can already do things better than humans. For example, machines, the new computers, put in parallels, can solve differential equations infinitely faster than human beings can do. Now, whether you're going to call that a case of surpassing human intelligence, I don't know. It certainly can do things faster. There are respects in which computers just don't even come close to human beings. So I'm inclined to say or view the whole question of, will computers ultimately surpass human beings' intelligence not to be a clear question. As I've suggested, if you look at intelligence in one way, they're not. So it's not a well formed question for me. Lloyd (U.C. Berkeley): I will give you a typical philosopher's answer for this. I depends in what you mean by intelligence. Already computers are able to do certain tasks which we take to be cognitive tasks much better than we can. I don't think this makes them more intelligent. I think that there are many kinds of human intelligence. There is artistic intelligence, there is mathematical intelligence, there is a kind of verbal ability, there is the ability to see the whole picture, the ability to see both sides of an issue. There are just so many aspects of human intelligence which are vital. I don't see artificially constructed machines as being able to perform all of the functions which we would naturally attribute to human intelligence. I do think that machines will be able to surpass us on some of these tasks, but not on intelligence per say, not on intelligence overall. Matson (U.C. Berkeley): Noat any rate not using any conceivable refinement of the Turing machine (digital computer). Turing machines necessarily follow contextfree algorithms; that is not the way we think. McCann (University of Southern California): I guess it would depend on what factors you have in mind. In terms of calculating lots of big columns of numbers quickly, obviously computers can do that. Although of course whether they are actually computing or calculating as opposed to what is really happening, a bunch of electrical states flip-flopping inside the machine with the results being interpreted in certain ways. The thing that stands most in the way of getting a straightforward yes or no answer is just that I think we do not have much of a hint of what human intelligence is. I am very persuaded by Howard Gardener's work on multiple intelligence. He is claiming that the sorts of capabilities for doing certain tasks quickly, that we called intelligence, get measured on the standard intelligence test, and things like that in our culture are just a very narrow range of human competencies that are artificially selected out or artificially highlighted. In a South Pacific's Island's culture, for example, th ability to navigate by stars is a crucial part of intelligence, but it is not exactly noted by us. In fact, I think there is a big indeterminacy in the notion of intelligence, whether human or artificial. And then once you go on to say what are the comparisons and contrast of human intelligence and artificial intelligence the questions are sort of fatally infected with the multiple ambiguities involved in the notion of intelligence in general. There is no doubt that machines can do some things that we count as intelligent tasks better than we can, but there is no doubt that there are a lot of things that we can do with the results of some of these processes that the machines cannot and maybe would not be able to do. McGray (University of San Diego): The answer is "yes" and "no." Computers are much more adept than we are at certain kinds of consistency tests and certain kinds of expert systems. But some other sorts of questions, even some simple problems in first order predicate logic, cannot be decided by any machine. Pippin (U.C. San Diego): The real question is a philosophical one: What is human intelligence? Rosenberg (U.C. Riverside): Yes, because human intelligence is the result of the operations of a machine. There is no reason why better machines can't be made. Ross (Claremont Scripps College): No, because a computer needs a programmer to teach it what it needs to know. A computer doesn't have an imagination and without an imagination the computer will be unable to form questions and answers itself or ways of solving problems. A programmer has to do these things, so a computer will always be dependent on programmers. Roth (Claremont McKenna College): That's a really interesting question. The first response I would give I think if we are thinking about human intelligence at it's best, my guess is that artificial intelligence will not be capable of surpassing or even equalling human intelligence, especially if we look at the subtlety and the kind of nuances, the imaginative potential that there is of human intelligence. I'm looking more on the side of creativity. On the side of our intelligence that is laced with feeling, with aesthetic qualities, things of this kind and it seems to me to forget it at least as I'm sitting here now. That it would be difficult to imagine that we could artificially create something that would be equal to that kind of subtlety in terms of intelligence. The other part is a little fictitious, but not entirely so. If human beings fail to develop the potential of their own intelligence it's conceivable to me that we might create artificial intelligence that would be superior to ours. We might be more rational in some ways. So I think this is another thing, the human intelligence is not a fixed element; it's something that could become better or worse, depending on what we choose to do with it, how we develop it. Sometimes we are not nearly as intelligent as we think we are, or as we could be, but I guess I'm impressed by when I look at what the human mind has been capable of doing. That it seems to have a range and a scope on one hand and also a subtle dimension of creativity that I find it hard to define. Schwyzer (U.C. Santa Barbara): It's such a frightening concept. Intelligence by itself is not very interesting. I think that some human should go along with that intelligence. It makes no sense to just have intelligence and nothing more. It's like having weight without size. We can have machines, but intelligence is a human attribute. I suppose I am a humanist. I fail to be fascinated with non-human things. I do have a computer in my office; however, it hasn't been used yet. It's good decoration. Shalinsky (U.C. San Diego): It's not precisely clear what this question is asking. To claim that some intelligence "equals" or "bypasses" another is quite vague. It is certainly clear that many forms of artificial intelligence surpass human intelligence: the calculator can perform functions that humans can not, a plain old digital computer can perform functions that humans cannot, and a super computer can perform functions that humans could not even dream of performing. The question ought to be phrased differently: will artificial intelligence equal or bypass human intelligence in the realms in which the latter is now superior? This is an important question, because it applies an important distinction between the kinds of intelligence behavior (e.g., numbercrunching) better performed by a very fast computer processing in serial fashion, and the kinds of intelligent behavior which are best performed by parallel processors. Humans process in parallel, and this accounts for their ability to perform and unders and in complicated contents. Currently, for example, the prospect of writing a computer program which will model even the simplest kinds of human behavior is quite dim. Consider, for example, the human capacity to interpret utterances: while we understand the meaning of "Mr. Smith watched the fireworks go up in his pajamas last July 4th," the computer has considerable difficulty. While we manage to recognize even as many as hundreds of different faces, the computer has considerable [trouble with face recognition]. In my view, there is absolutely no reason to think that parallel computers will not equal intelligence (even in the domains in which the latter currently surpasses the most advanced artificial intelligence)but this is just a bet, after all! . Sircello (U.C. Irvine): Machines will be able to do more, but will not be more intelligent than humans. Suppes (Stanford University): Already in certain respects, of course, computers could do things better than human beings. For example, computation. Other things they that can't do as well. So I think what will happen will be an increasing complicated comparison. Computers will continue to acquire capabilities they don't now have and so the comparison and kinds of tasks they could do, how well they do in comparison to, show how well humans do, will continue to change. Wollheim (U.C. Davis): The problem that confronts us first is how to introduce consciousness and meaning. I don't have any conviction that this can be done. Woodruff (U.C. Irvine): I think that hardware, as well as wetware, as it's called, can do these things, in principle. So artificial intelligence of this sort is possible. I think to some extent it already exists. But what people have in mind, I suppose, when they ask this question is that they imagine it being like an alien coming and saying these things that we just can't understand and saying, "Oh, he's so smart, he's smarter than we could ever. . ." and so on. Do I think that sort of thing will happen? I can imagine it happening. I think that for a long time, humans would understand how it happened. That is that one would have to create some kind of quasievolutionary mechanisms that would allow machines to evolve, so they could go beyond humans actually programming them. Although even now, computers have certain abilities that even though we program them to do these things, what they do is so complex (because it's so large scale) there is a certain sense in which we can't understand what they're doing. I think that we can understand what artificial intelligence is, and that it's not, in principle, different in kind from what humans do. If there are differences, they have to do with the fact that we are different kinds of machines than electronic computers. We are massively parallel, and we have all these interconnections in the brain which people are now trying to understand, stuff called neural net computing. But it's not any kind of ontological difference, not different kinds of stuff or substance in mind and matter. So do I think that artificial intelligence will equal or bypass human intelligence? I'm not sure, but I certainly wouldn't be surprised. The reason I wouldn't be surprised is that we create machines that are much more powerful physically. I don't see any reason why we cannot create thinking machines that are more powerful than we are. In fact, we've already done it in certain respects. I especially don't think that human intelligence is something that is es entially different from machine intelligence. Our brains thinking or electronics thinking are essentially the same thing.