Scanned image from
Mark Harden
If there is anything worse than a tool that doesn't work, its a tool that works poorly. If you try to use a tool that is totally useless, you quickly throw it away in disgust and try to find a better tool for the job. A tool that sorta works is seductive. If you twiddle with it for a while you might get better at using it and if you persist it might even get you through the job, after a lot of effort. You may never get around to looking for a better tool. The history of science is full of such tools that worked poorly. Models of the solar system that placed the Earth at the center are a popular example. By twiddling with such models, say adding epicycles, a reasonably good fit between the predictions of the model and the actual observations of planetary movements could be achieved. Eventually, it took a conceptual revolution to replace the baroque system of epicycles with Kepler's more elegant model involving heliocentricity and ellipses. Programmable digital computers are the darling tools of our age, but they too have their limits.
Computers have come to play an important role in the philosophy of mind. At first, the only computers we had were people who did numerical calculations. Since people are generally poor at computation, there have always been attempts to find tools to aid in computation. The electronic digital computer was a revolutionary advance in machine assisted computation. General purpose computing devices that can execute an unlimited set of programs (each program itself being sets of instructions specifying computational algorithms) are very powerful computers. Just how powerful they are is an interesting issue, an issue that Searle and Dennett have argued long and hard over.
In general, it is thought that general purpose digital computers are universal Turing machines (for what is still the best description of Turing machines and their implications for MIND, see Godel, Escher, Bach: An Eternal Golden Braid by Douglas Hofstadter), capable, in theory, of running any program that can be set up as a well defined algorithm. An algorithm is just a specific set of instructions for how to perform a specific task. Of course, it is often not a trivial problem to figure out how to solve some tasks by computation, algorithms, or a computer program. Artificial Intelligence (AI) can be described as a branch of computer science concerned with figuring out how to get computers to do tasks that people can do. AI has had many successes. It was fairly easy to program computers to never lose a game of tic tac toe, so easy that this is a common exercise for beginning computer programmers. A no-lose algorithm that is learned by five year olds goes like this: always block you opponent and if you can, take the center square first, then the corner squares. It took a little more effort to find an algorithm that would turn a digital computer into a champion checkers player (see Samule's program in Computers and Thought). It took about fifty years and spiffier computing devices to make a world-class chess program (Deep Blue). There are still many things that people can do that computers have not yet been programmed how to do.
A philosophical debate has arisen: are there things that people can do that programmable computers cannot do? It should be clear that a key issue in this debate must be distinguishing between what it is simply hard to get a computer to do and what is impossible to get a computer to do. In the case of the argument between Searle and Dennett, Searle thinks that programmable computers can do less than what Dennett thinks they can do. The debate is simplified in this case because even Searle admits that people can be viewed as being a type of machine that can produce a mind and do all of the amazing things people can do. So the question becomes, are there some mechanical processes (like human language understanding) that humans can do that digital computers cannot be programmed to do? Searle says, "Yes!" Dennett says, "No!"
Who is wrong? Dennett admits that his "no" is a kind of theoretical "no", not really a practical "no". He does not know how to program a computer to understand language, but he is sure that it is theoretically possible. Thus Dennett argues that programming a computer to understand human language is a hard problem, much harder than the chess problem that took fifty years of effort by AI researchers, and which really was not solved in an elegant way, only by brute force. It might take AI researchers much more than fifty years to solve the language problem, Dennett will not even venture a guess. In fact, when this argument between Searle and Dennett started, Dennett was sure that there would have to be major improvements in the tool box of AI research in order to even be able to make serious progress on the problem of language. Dennett admits (page 367 of The Mind's I ) that a computer that only manipulates formal symbols will be hard (actually, Dennett sometimes even implies that it would be impossible) to get to understand language in the sense that humans understand language. Dennett argues that we need to add new tricks to the AI tool box in order to really make a machine that will think and use human language. This really means that for all practical purposes Dennett agrees with Searle's claim that digital computers as they now exist within the world of AI research are, for all practical purposes, inadequate for dealing with some tasks that humans can accomplish.
The qualifier, "for all practical purposes" is a sticky issue, the issue that has generated most of the heat between Searle and Dennett in their argument over the power of Good Old Fashioned AI. Dennett cannot tell us how to program a machine to use human language other than to say, "Make a human baby and let it grow up exposed to other humans who use language." However, Dennett can argue that it is theoretically possible that some day a program for a digital computer might be found that would allow a digital computer to understand language. However, using the Von Neumann architecture, such a program would probably be too slow to actually converse with us in real time. So if we want to make a computer we can chat with, we need to expand the tools available to Good Old Fashioned AI and make computers that are more like biological brains.
Thus, a key aspect of New-fanggled AI has become how to get computers
to learn tricky things like how to use language. Most computer programmers
quickly recognize the difficulty of programming a computer to understand
language, but can we program a computer system that will then be able to
learn to use language on its own? There was much excitement in the 1980's
when some new tricks were found for making it easier to train computers
to solve certain problems. The tools of so-called "neural nets" and parallel
distributed processing" (PDP) were brought into AI research and they were
certainly important advances. For a crude estimate, I would guess that
the methods of Good Old Fashioned AI would have required about 1,000,000
years for construction of a computer that could understand human language,
and even then that computer would have been too slow to carry on a conversation
in real time. With the methods of AI by PDP, I guess that it would take
several centuries to make a computer that can think, and it would be a
slow talker, although even its impatient critics would live long enough
to have to admit that it worked. I would not mind seeing true human-quality
AI before I die. If I am to see a man-made intelligent device it will be
because more of the tricks used by biological brains are imported into
the tool box of AI.
After Dennett and Searle die they face judgment together........
St. Pete: I brought you both here today because your argument about Strong AI and the Mind has raised doubts about whether you both lived your lives as noble philosophers, worthy of final rest in Heaven, or as bickering ignoramuses who deserve to go to Hell.
Dennett: "Whew! I had feared that there was a special Hell reserved for all philosophers."
St. Pete: "No, but there is a special Hell for each individual, and it is my job to sometimes make sure that people get to experience their personal Hell."
Searle: "Yikes! I must confess that debating with Dennett about Strong AI often seemed like Hell on Earth. I hope that we are not here together in the After Life because we are going to continue to share that Hell for the rest of Eternity."
St. Pete: "I sense that you are both troubled over the prospect of going to Hell and unable to maintain, shall I say, a philosophical attitude about the prospect. I can tell you this, He let slip the fact that although it may be true that you both deserve to go to Hell, since you did debate a yes-or-no philosophical issue and one of you had to be correct, only one of you will go to Hell.
Dennett: "You should have brought your asbestos nappies, John, I'm sure I was right about AI."
St. Pete: "Better not count your chickens before they hatch, Dan."
Searle: "Why keep us in suspense? Just tell us who was right about AI and which of us is going to Hell."
St. Pete: "He knows which of you is going to Hell, but He has cornered the market on omnipotence. The three of us will find out together which of you is going to Hell."
Dennett: "How will you decide the case?"
St. Pete: "Again you jump to an unwarranted conclusion. There is a decision algorithm, but all I do is execute the program."
Searle: "Are you saying that some computer is going to decide which of us goes to Hell?"
St. Pete: "Well, 'some computer' is rather vague language for a philosopher to use, but rather than pick nits, as I recall from your writings, both of you agree that the universe is a type of computer."
Dennett: "And if you are part of the universe then you are a computer too, so it could both be true that you are going to decide and a computer (you) is going to decide."
St. Pete: "Well, I wouldn't be able to sleep nights if I had to have it on my conscience that I was the one that decides who goes to Hell, so the process is a little less personal with respect to my role. I have here....." [St. Pete unrolls a glowing golden scroll] "......the algorithm for deciding which of you will go to Hell. The instantiation of this program is all His handiwork, just like everything else.
Searle: "Wow! So God is a computer programmer, eh?"
St. Pete: "In a sense. Shall we get on with it then?"
Dennett: "Can't you show us the code? What language does God program in?"
St. Pete: "Quickly now, I'm a busy angel." [St. Pete turns the scroll around and lets the two philosophers have a look] "I'm afraid it will only appear as a meaningless collection of symbols to you, but I'll tell you what it means. [St. Pete turns the scroll around again and reads] "In a minute I am going to place each of you into your own personal testing room. In your room, you will find an Intelligent Robot."
Searle: "A real functioning robot that can think and talk?"
St. Pete: "Yes."
Dennett: "Oh oh, John, its looking bad for you! Real thinking robots!"
St. Pete: "Dan, please stop jumping to wild conclusions."
Searle: "I've always said the people are just thinking machines. The question is, are the minds of these robots powered by Good Old Fashioned AI?"
St. Pete: "Hmmm......." [St. Pete mumbles as he reads the entire scroll to himself] "It doesn't say if the robots were created by Good Old Fashioned AI."
Dennett: "Can you at least tell us if it is possible to make a thinking robot using Good Old Fashioned AI?"
St. Pete: "No, the decision algorithm explicitly precludes me from answering that question. However, when you get into your testing room you will each be allowed to ask your robot one question. Further more, your robot knows the answer to every philosophical question that you have ever asked."
Searle: "So I can ask the robot about AI?
St. Pete: "Yes. Are you ready?" [Searle nods and is whisked away to his testing room]
Dennett: "Yes." [Dennett is transported to his testing room. After 10 seconds Searle returns to St. Pete]
St. Pete: "Well, how did it go?"
Searle: "Well, I got to the testing room and I asked the robot, 'How can I convince Dennett that he was wrong and that claims of Strong AI were wrong?' The answer from the robot was, "You can't. I am an example of a computer running a conventional AI program which creates my mind.' Then I came back here."
St. Pete: "Don't let it bother you. There are worse things than being wrong. Let's go. [St. Pete starts pulling Searle up towards Heaven.]
Searle: "Can't we wait for Dennett, I'd like to hear what he asked his robot."
St. Pete: "I can tell you that. [St. Pete checks the golden scroll.] Dennett's question was, 'How can I convince Searle that he was wrong, tell me how Good Old Fashioned AI can program a computer to have a human mind.' Now, let's go."
Searle: "Gee, I'd like to hear what Dennett finds out."
St. Pete: "I'm a busy angel, I don't have time to wait for Dennett to
come out his testing room. Its going to take most of the rest of eternity
for that robot to explain to Dennett how to program a computer to have
a mind. [And with that, Searle is taken to Heaven where he quickly gets
over having been wrong about Strong AI.
Go to John's Book Page.
Go to John's
Home Page.