The following appeared in Volume 98, Number 1 (Fall, 1998) of the APA Newsletters

Selmer Bringsjord, gave the keynote address, The Impact of Computing on Epistemology: Knowing Godel’s Mind Through Computation, at the 1998 Computers and Philosophy Conference held in conjunction with the XX World Congress of Philosophy in Boston. A Full Professor at Rensselaer Polytechnic Institute, he teaches in the departments of Philosophy, Psychology & Cognitive Science, and Computer Science. He is also director of RPI’s Minds and Machines Laboratory. Professor Bringsjord is definitely a person doing interesting things with computers, as the interest shown in his work by the popular press attests. Recent celebrity has arisen from the development of Brutus, a program that generates original short story texts. Brutus.1 was recently discussed in a radio interview with Professor Bringsjord conducted by Weekend Edition Saturday host Scott Simon and aired on May 2, 1998. That interview (including a reading of the short story "Betrayal") can be heard at the NPR website at http://www.npr.org. Search the site using "Brutus" as keyword . A RealAudio plugin is required. Links to articles by and about Professor Bringsjord are available at his website at http://www.rpi.edu/~brings/.

The following is an excerpt of an interview with Professor Bringsjord conducted by Philosophy and Computers Associate Editor, Bill Uzgalis.

Bill Uzgallis (B): Looking at your website I get the impression that you are involved with one of the most centrally interesting questions that computers pose to philosophers - in what ways people are and are not like computers.

Selmer Bringsjord (S): Yep. I found that interesting since an undergraduate studying philosophy. Absolutely.

B: I think that what computers are doing for us is pushing that question in the sense that as computers succeed in doing things that we didn’t think they could do and that humans could do it narrows the focus on what is uniquely human.

S: I think that way of looking at it is true for many, particularly lay people and to some degree scientists. Some scientists and philosophers predicted that a machine like Deep Blue would never arrive, but of course it did. But generally I think the question hasn’t changed and the insights that we have into the question whether we are machines or not essentially hasn’t changed. The arguments that I take to be key, and in fact in my case to settle the issue, are not affected by advances in technology, but I recognize that is not the attitude of other groups - say the man on the street. See, I believe that I know that minds aren’t machines, so one of my strategies to try and convince people that that’s true is always to keep an eye on the advance of computers as they "get smarter and smarter" and then to debunk that smartness and also to try myself to play in that game - to try to get machines to

be more and more intelligent. I have no problem saying they’re intelligent - if that term is cashed out in some behavioral fashion. They are intelligent, they’ve been intelligent for a long while. Deep Blue is intelligent. But I think we have all the information we need going back quite a ways to establish that they are not ever going to be people. Behaviorally they are already animals, simple animals, and they will become very sophisticated artificial animals, in my opinion.

B: OK, let’s sort of flesh out your disagreement here with the notion that minds are machines. And I mean one line someone might go down is to say, ‘well, there are all these things that people claim that computers can’t do, and slowly, or perhaps quickly, some of them are toppling’. Then there are things like the Chinese Room argument which goes to show that in principle the human mind is different from a digital computer, and I take it to some degree you agree with Searle.

S: I do. There’s a chapter in my book What Robots Can and Can’t Be, called "Searle", where I simply try to modernize the case, try to inoculate it against some objections he didn’t, at least at that point, rebut. So I’m definitely a Searlean - Searlean through and through, and though I have a number of arguments that I have originated against computationalism, not only am I a Searlean: I think that argument is in some sense more fundamental than many others, including some of the other new ones that I’ve given. As I think I mentioned yesterday in my talk [CAP keynote, .ed] I constantly see Searle’s ghost popping up in other arguments and in some cases those are my arguments against computationalism. People criticize him sometimes for using slogans, however in this case it’s a slogan that sticks, I think it’s true, and again does creep around the rest of the ball game in philosophy of the mind. And the slogan simply is: "Computers move symbols around, we don’t." And in the case of a machine proof of Godel’s first and second completeness theorems at least done by the people I’ve looked at, that slogan is borne out in striking fashion - it is just moving symbols around, in fact so starkly, that even computationalists I think, when they look at these proofs, back off from the proud declaration made by a number of people that the machine here has done something - on par or nearly on par with a person, with a mathematician. But there are other arguments I think that are just as fundamental, that are just as good. There’s an argument in What Robots Can and Can’t Be from introspection from a weak form of incorrigiblism, which to my knowledge is new. No one had given it before. Only two people have ever reacted to it - at least its technical side, Bill Rapaport, at SUNY Buffalo—- is one of those people. There’s a dialogue between he and I and Jim Fetzer coming out in, maybe it’s already out, in Journal of Experimental and Theoretical AI. And that argument is the one I’m really most proud of. And you know I’m partial. I think its a sound argument. But sure I’m a Searlean, definitely. Not only with respect to the Chinese Room. The Discovery of the Mind is a book that also contains a lot of things I agree with. His more recent argument that computation is well nigh a meaningless term because it seems to apply to everything including walls and pens —— is an argument that I think is pretty powerful. I’ve grappled with it and I don’t think it works exactly the way he says it does but there’s something there.

S: Well, there are a number of things in your question here to comment on. There is a free-will argument which I give in What Robots Can and Can’t Be . There is a chapter devoted to it. I can’t say that I define what free-will is. I end up giving the fairly standard argument for the libertarian view of human action. And then I try to establish that that view of the person is true. And then I try to show that if it is true, we cannot possibly be Turing machines or any equivalent.

B: Well, it seems clear that if the libertarian view of free-will is true then machines are not going to have free-will. I think that is absolutely right. But if you are a compatibilist, then it would seem quite possible for a computer to have free-will, at least in principle.

S: Oh yah! I tried to face up to the challenge in the chapter of giving an argument for libertarianism. It wasn’t just a conditional argument—if libertarianism is true then computationalism is false. It is an argument starting from the so called Dilemma, a variation of the Dilemma going back to Roderick Chisholm for why libertarianism is true. So, that is in there. Consciousness, well there, I don’t know if we will ever be able to say what it is in any third person terminology. I take that to be part of what Nagel originally argued for (if not that, then pointed toward). So, I don’t think there is anything to be done if you are an AInick and you feel that your robot needs consciousness to get something done. That itself is a difficult question, by the way, what role it could play for a roboticist. But suppose you were convinced of that. I think you would have a devil of a time even figuring out where to start. You might even draw a blank on what you are going to tell your research group about building in this thing called consciousness. You wouldn’t have a clue. I think it is no accident that in the robots we see, cutting edge robots, not only is their behavior such that we are unwilling to ascribe consciousness to them, but when we look at the details of their design, and the algorithms and the planning that went into their construction, there is nothing, even in that material, which suggests a conscious attempt to build in consciousness. There isn’t anything to suggest that the designer, the builder, the team, knows what consciousness is. We have no idea.

On the issue of what it [consciousness] could bring a robot, an artificial agent, there I think there are some very good answers now in the literature. Some of them I have given. It is again related to Searle. There are things which don’t show up at the level of mere animals, but there are things which show up at the level of people, especially people who try to be creative. I think you have a very hard time originating anything, as Lady Lovelace said. She said that was the problem for the Turing Test. You have a very hard time originating anything, I think, unless you have phenomenal consciousness, subjective awareness. My favorite example is the writing of fiction. Yes, it is possible that there could be automatic creative writing, done in a trance or something, but even there there might be phenomenal consciousness. Generally, if we leave out bizarre cases like that, an author has to capitalize on his or her ability to feel what it is like to ——-. Now it might be what it is like to be a certain character, or what it is like to be in a certain place. So, there, I think it may show up in practical terms in AI. The absence of a machine that can do anything creative, maybe due to the fact that these machines are only animals or zombie animals.

B: Well, let’s consider creativity. It seems to me that if you ask what is creativity, there are two basic models. One is the creatio ex nihilo —that is what is created has to be something absolutely new. And in that sense I think this discussion parallels the free-will discussion. This view of creativity is kind of like the libertarian view of free-will. If it is true, then computers cannot be creative. On the other hand, that view seems to be completly implausible. What seems to me to be true about creativity is that a better model for it is the recombination of already existing elements, perhaps to serve some new purpose.

B: Well, tell me about Brutus in this regard. You have been teaching, in effect, or creating a program, which writes short stories. And to what end?

S: Yes. Not teaching, because it’s definitely not AI work that subscribes to any of the tenets of connectionism or variants thereof. I think that approach to AI is doomed when it comes to the attempt to build persons. It is invaluable when it comes to building aspects of persons, and it is the key to the problem when it comes to building animals or things like animals, artificial animals. But it is never going to work when it comes to high level intellectual behavior. In my opinion, that kind of behavior is just irreducibly symbolic. And I would include the writing of novels there. So, it is not that we are teaching it. We are constructing a logic machine that is able to generate short stories. Yes, as you said. Brutus.1 does not originate anything. Brutus gets knowledge by injection. To give it knowledge of betrayal, you figure out how to formalize betrayal and you put the formal account into the system. With self-deception we took Mele’s account basically off the shelf. We did the stuff on betrayal in classic analytic philosophical form where you come up with a definition, try a counter-example, and try to evolve it—the stuff that I was trained to do at Brown with Roderick Chisholm. There are many reasons for building Brutus.1. It is much more interesting to me than a lot of AI work. I would rather tackle extremely difficult AI problems. Although I am happy to do the other stuff, I would rather tackle the hardest of the hard problems because it is more philosophically interesting. It is much more philosophically interesting to try to build a machine able to create literary artifacts, in my opinion, than it is to create a machine able to mechanically verify a proof. Machine verification of a proof, though I am happy to work with those systems, I won’t invest any real heart and soul in the research. But I will in the area of creativity. It is also a way of constantly showing my opponents that I know the state of the art inside and out. And I know when someone says: "You are not a computationalist and you are wrong because of this system"—that it is a fake—that they cannot be serious, at least so far, as the century is about to turn. This is because I know all the systems that are supposed to be creative in AI, from the overwhelmingly honest ones that Hofstadter has come up with (They are very honest because he has limited the scope tremendously in his attempt to try to figure out what creativity is. They are honest because they are not oversold or overbilled.) to systems which are fakes. And there have been fakes in the area of literary creativity. This may sound a bit harsh here. But now and then the media carries stories about systems which have written novels. Systems basically can’t write novels. There are expert systems which will help you sort out how you ought to write a novel. And now some of these have made it into the marketplace. If you are a script writer, you can now buy these. They are not even that expensive anymore. You know, $500 gets you a good system for helping you write a script for the movies. That technology is good but it relates to some things which have been oversold. Deep Blue, that is absolutely the worst thing that could have happened to AI, if you are a true believer. John Mcarthy picked up on this early on. Deep Blue is nothing. It has done nothing that would fundamentally gladden the heart of a strong computationalist.

B: Why is that?

S: Well, because it does what we knew to always be true, if we simply looked at the math. Chess is a solvable problem. I don’t have much interest in devoting research time and energy when it comes to constructive AI to solvable problems. If a problem is solvable in the ordinary sense of theoretical computer science, we know a Turing machine can do it. And then if it is tractable, it is an even bigger "So what?". It is tractable, so you can build a system that can do it. We knew that about chess before we ever started this nonsense about whether the best human chess player could be beaten. It is amazing to me that Dreyfus or anyone like that (forget predictions that it wouldn’t happen) could ever have thought that it might not happen. It boggles my mind that anyone could ever have said that. If the problem is solvable, and there is even a whiff that it is tractable, it will happen unless the human race goes extinct. It will happen. Raw computing power is increasing every single minute. As we sit here, the prices are dropping because the chips are getting more powerful. RPI, my own University, is one of two places selected by the government to run another version of Sematech. RPI with the University of Albany with a partnering with Georgia Tech and MIT to build the next generation of hardware. This stuff goes on and on and on and on. And it won’t be too long before in tournament play it is a completely done deal. The machine will beat any grandmaster, any master, on any occasion for playing chess in a competitive situation. It is just going to happen.

S: That’s right. Absolutely. Some people classify genuine AI problems according to whether or not they are tractable, knowing that they are solvable. There is an entire, and mature branch of computability theory, complexity theory, concerned with issues of tractability and efficiency. That’s still bad, that is not the mark of a genuine AI problem. A lot of the field doesn’t go along with this because the field countenances an expert system able to do a relatively simple medical diagnosis as a true AI system. I don’t. The hallmark is whether it is an open question as to whether it is solvable or not. So, this chess thing was never an open question. It might be an open question whether a system can write on the level of ordinary journalism. But it is certainly an open question at the level of belletristic short story writing. And it is certainly an open question at the level of a novel. That, therefore, is something worth a philosopher’s time.

B: Right. So how is Brutus doing in that regard?

S: Well, Brutus is doing very badly, I guess, from my standpoint. But it was never intended to do well there. It does very well as a system engineered by hook or by crook to appear to be a system capable of genuine literary creativity. There is a difference between a system which can pass the Turing test and a system which can genuinely think, if thinking is cashed out in terms of phenomenal consciousness. So, Brutus does well from the standpoint of the Turing test, and I hope will do a lot better in the future. But it does awful from the standpoint of raw creativity, real creativity.

B: That reminds me, when I was looking at your web site, I noticed that you had different levels of Turing tests. There is the original Turing test, and then a body that can fool you into thinking it is a human body, and so forth. It seems to me that at some point or another these stop being Turing tests and become something else. What Turing was all about was behavior, linguistic behavior primarily, but other behavior as well. When you get to brain scans you seem to be checking not whether behavior is similar, but whether the structure of the brain is similar.

S: Yes, that may be true. This sequence of ever more stringent tests is due originally to Steven Harnad. I don’t know, now that I think about it, why we persist in calling them variations of the Turing test. I guess you are right. They are clearly based on observation and inference from observation of empirical stuff. I guess that is why we can’t resist the temptation. They are definitely more. But I suppose someone like Harnad would say, "Yes, but Turing shouldn’t have restricted the fundamental idea to linguistic evidence which is the restriction of the original Turing test". So, I don’t know, I will have to think about that one.

B: Well, one of the things which is so impressive about Searle’s Chinese room is that it is not only an argument against the Turing test but against functionalism as well.

S: Yes, absolutely. That is another thing which astonishes me. There are philosophers of mind who don’t agree or don’t see that functionalism, at least the classical varieties, are not just overwhelmingly embarrassed by the Chinese room and by the arbitrary realization arguments. Even if you don’t think it works, you have to admit that it is a huge embarrassment. And it is really befuddling to students. In my experience, students who see the argument are really puzzled about why functionalism would still be the dominant view in the philosophy of mind. Granted I may put some nice slants on it, and give the arguments forcefully, they can’t figure it out. Why is this still the dominant theory of the mind? Maybe there shouldn’t be any theory which anybody affirms with any level of confidence. They would be happy with that. But, Searle’s argument certainly does that.

B: The reason I mention this is related to the point about levels of Turing tests. The original Turing test has associated with it a rather weak criterion of intelligence. Functionalism has a much stronger criterion of intelligence. So, the Chinese room argument really raises the stakes in significant ways.

S: It does. It does, and I suppose that is why it is quite irrepressible. It does not die, as much as many people in AI would like to see it die. They themselves, find themselves talking about it and dealing with it. They may even take vows that they are not going to discuss it at the late night sessions at AI conferences. But, in my experience, invariably it pops back up. Again to relate back to something we talked about earlier in our conversation it is because the slogan at the heart of it is not just a cheap slogan. He is really getting at something. That is why Searle can give an argument. I am much more than a formalist than he is when it comes to giving these arguments. But that is why he can give an informal argument and have it last, because there is something to it. And someone can investigate it and try to formalize it and improve it.

B: O.K. well, where are you going from here?

S: There is a book called Super Minds —super minds in the sense of non-computable cognition which will be coming out from Kluwer. I am working on a trilogy. The first one was What Robots Can and Cannot Be, the second one is Super Minds, about non-computable cognition. And the third book is much more practical, about how you would do engineering in AI, having embraced the view that the mind is capable of processing uncomputable functions. So, Super Minds as a manuscript is in its last stages of revision. So I will probably finish it up by the end of this academic year. So, that would be Spring of ’99. The book on Brutus.1 and literary creativity is going to Lawrence Erlbaum. It is called AI and Literary Activity: Inside the mind of Brutus I, a story telling machine. That is going for copy editing on October 1st. So, these books will be in the pipeline right about the same time. They will hit the shelves about the same time. I regard them as essentially done, and though I love them they are no longer taking any creative energy. There is a dialogue I am working on. I wrote a small book for Hackett on abortion. I am working on a small dialogue for Hackett on human rationality that, even though it is designed to be a dialogue, is new material for me. These are new arguments. There was a new argument in the abortion book, that is how I ended up writing it. It was a new abortion argument which I came up with through the years and made it the anchor for the book, but it was very closely related to the literature. This stuff in human rationality is new. So, I am working on the first three chapters of that. I don’t have a contract yet, but I think I will sign one with Hackett. I continue to work on a book Godel and the Mind which is a large book and a lot of parts of that have been written. Most of this I am not publishing in journal form because I want to save it all for the book. That is a big project for me and a really special project. That includes the material that I presented at the keynote for the CAP conference here in Boston on the machine based version of the incompleteness theorems. That is a chapter in the book. There is a chapter in the book devoted to how smart these machines are that do the proofs mechanically. ‘Not very’, I say. I give complete coverage in a monstrous chapter of the Godelian argument against the computational conception of mind which in its published versions is awful. That includes Penrose’s version in Shadows of the Mind —which is the most recent—which, for formal reasons, is awful.

B: Who is the first person who gave that argument? It is an old argument.

S: It is an old argument. I think the first was Lucas. I am no a scholar, however, and I would not invest money, wouldn’t wager that this is the case. There had to have been people who were thinking about it before Lucas, but Lucas gets credit for originating it. It is a terrible version. I do think there is a victorious version which I hinted at in What Robots Can and Can’t Be in my chapter on Godel. I think there is a modal version of the argument which maybe I should publish because somebody else will think of it. I do think there is a victorious version of the argument which is much harder than modalized versions of the argument. But I think Lucas is the first.

And then I am working on a textbook. I no longer use logic textbooks in my logic classes. I use my own textbook in progress. That is interesting because it is not really all that creative but it is really enjoyable to fully assimilate the range of proofs for a long established theorem, and then to turn it over and over in your mind to try to find a variation or a more attractive presentation. It is kind of weird, because it is not that creative, but I love it so much that I don’t much care whether the textbook ends up coming out and making me money, I just find it aesthetically pleasing. So, those are basically the things which I am working on in terms of books.