I like to think that I am re-posting all of John's blog as some sort of service to humanity, but really I just enjoy rediscovering gems like this one. John's review of a biography of Kurt Gödel has been definitive in shaping my opinions about AI and computation.
In short, I don't think strong AI is possible, and this is the true explanation why computer scientists have spent the last seventy years looking for it without finding it.
Roger Penrose famously criticized strong AI in his book The Emperor's New Mind. Wikipedia's summary claims that so many eminent scientists have criticized Penrose's position that it is effectively refuted, to which one might reply, "OK, when where are all the AIs?"
I think Penrose truly fails by looking for the mind in physics. He is really just embodying the spirit of the age, but it is a sad thing to see when he was perceptive enough to notice that the essence of thinking, abstraction, is not algorithmic.
There really is a similarity between what minds do and what computers do, but the real similarity makes AI less likely instead of more. Computers are the instantiation of the immaterial forms of Plato [thereby proving Aristotle right]. Ross's conference presentation also illustrates the dangers of treading outside one's field. I do it, I like to do it, but I am always aware that I can sound just as silly to others as they sometimes sound to me. Ross makes an off-hand comment in his presentation about the deadliness of dioxin, which was quite the trendy toxin for a while. Then the Russians tried to poison the Ukrainian presidential candidate Viktor Yushchenko wtih dioxin [a preview of the recent unpleasantness], presumably under the impression that it was exceptionally deadly, only to find that all it did was give him a bad case of acne. Oops. Maybe that was just a clever counter-intelligence ploy in the wilderness of mirrors, like the time we sold the Russians faulty gas equipment.
I'm not an expert in toxicology, but I at least need to know enough to be able to accurately communicate with the experts so I can demonstrate the products I design are safe. Dioxin isn't nice stuff, but the dangers were wildly overblown.
This was also the beginning of the end of my interest in Neal Stephenson's books. His environmental thriller Zodiac featured a plucky band of environmental crusaders who thwarted a plot to dump dioxin in Boston Harbor. I already knew that dioxin wasn't all it was cracked up to be, and once I noticed one that was a little off, I started to notice a lot of things that were a little off. Oh well.
A Life of Logic
by John Casti and Werner DePauli
Perseus Publishing, 2000
210 Pages, US$25
Kurt Gödel (1906-1978) was the mathematician and logician whose now famous incompleteness theorem easily ranks among the most uncanny products of the notoriously uncanny first half of the European 20th century. This very brief book by two computer scientists does try to fit Gödel into the world of scientific Vienna in the1920s and 30s. (The book started life as a program for Austrian television: there is a great deal of talk about mysteriously undecidable recipes for Sachertorte pastry.) The authors are more concerned, however, to explain the theorem itself, its relationship to the idea of computability, and the connection all these things have to such questions as the feasibility of artificial intelligence and time travel. This is an unmanageable amount of ground to cover, and the treatment is uneven. Still, simply addressing all these topics between two covers is an accomplishment. The authors provide a blessedly brief, ten-item reading list for those who want to look more deeply into the separate areas covered.
Gödel was born in the town of Brno, in what is now the Czech Republic, to a family that had grown wealthy from textile manufacturing. The Gödels were German-speaking. The authors tell us they were not Jewish, but we learn no more about confessional affiliation, beyond the fact Kurt was anti-Catholic all his life. Gödel entered the University of Vienna to study physics, but switched to mathematics after a few years. He soon became a member of the Vienna Circle, the influential group that sought to reduce all philosophical questions to problems of language.
Like Karl Popper and Ludwig Wittgenstein, who were more loosely associated with the Circle, Gödel's membership probably helped him most by providing fodder for criticism. Indeed, few thinkers have ever been less interested than was Gödel in closing down metaphysics. If mathematical Platonism were a religion, Gödel would have been Billy Sunday, his American evangelist contemporary. For Gödel, mathematical objects were as "given" as lumber. They are just another kind of semantic content of sentences. What Gödel did in his proof, the first published version of which appeared in 1931, was to show the weakness of syntax, the system by which semantic content is ordered. The incompleteness theorem shows that there are propositions that we know to be true, but that are nevertheless logically unprovable. A slightly more rigorous formulation is that any logical system at least as complicated as arithmetic will be incomplete, because it will be able to produce statements that cannot be proven or dispoven within the terms of the system. The natural languge versions of the "Liar Paradox" are of this nature.
While Gödel was thinking these deep thoughts, the politics and economy of the German-speaking world were going to hell in a hand-basket. The failure of the Austrian bank, the Credit-Anstalt, in the same year as the publication of the theorem is usually blamed for blowing up the already stressed European financial system. Austria's First Republic, created when the Habsburg empire disintegrated after the First World War, collapsed into rule-by-decree in 1933. Nazi Germany annexed Austria in 1938. (This happened, it must be said, with the approval of most Austrians.) The Second World War began in 1939.
Gödel divided his time in those years between Vienna and the Institute for Advanced Studies at Princeton, New Jersey. The Institute, acting in large part under the influence of John von Neumann, served through most of the'30s as a haven for scientific refugees from Europe. Gödel was not neglected. He was offered and took several temporary appointments, but he kept going back to the University of Vienna. Although too unworldly to have ever engaged in politics, he did lose his license to lecture, the Privat Docent, because of his connection with the Vienna Circle, which the Nazis regarded as too Leftist and too Jewish. However, he applied for and actually received a new license as a Docent of the New Order. It was only in 1940, when it was apparent he would be drafted, that he left Austria for good. He and his wife traveled east by train across the Soviet Union, then to Japan, then to the West Coast of the United States, and then to Princeton. His wife, Adele, did not like New Jersey, but they stayed permanently.
There are many legends about Gödel's antics at Princeton. This book gives us only a few of the best-known ones, such as how Einstein himself had to help calm Gödel down when the latter went to take the oath of citizenship. (It seems that Gödel had found a logical flaw in the federal constitution that would permit the creation of a dictatorship, and he insisted on telling the judge.) The most surprising thing to me, however, was that Gödel was actually a conscientious faculty member. His flaw was that he tended to obsess about the work of any committee on which he sat.
Although Gödel continued to produce significant mathematical results during his time at Princeton, he was never again as productive as he had been at Vienna. (His wife called the Institute "an old-folks' home," and she may have had a point.) In any case, his interests turned increasingly to philosophy. Gödel famously constructed an ontological proof of the existence of God (he was a great admirer of Leibniz, who had a proof of the same type), and an independent proof of personal immortality. (Karl Popper had one of these too, by the way.) We are told that Gödel was also interested in "the occult," but are given no specifics.
Gödel was paranoid, convinced that someone was trying to poison him. He therefore always made a great fuss about eating. When he died of what his doctor called "malnourishment and inanition," he weighed just 60 pounds. On the other hand, he also suffered throughout his life from some obscure gastro-intestinal disorder, so it is possible that an underlying basis for this behavior was simply never diagnosed.
Why should we care about crazy old Kurt and his annoying theorem? For one thing, it's immensely practical. The theorem, and Alan Turing's related Halting Problem that was developed at about the same time, are key to our understanding of what computer programs can and cannot do.
Perhaps the most interesting such question, covered at length in this book, is whether it is possible to construct an artificial computer intelligence. In "The Emperor's New Mind" (1989), Roger Penrose revived an argument based on Gödel's theorem against the possibility of an algorithmic machine mind. To put it briefly, Penrose pointed out that people can spot "Gödel sentences," true but unprovable propositions, that computer programs cannot detect. Thus, he reasoned, whatever else the human mind is doing when it spots these sentences, it is not computing. Refutations of Penrose are usually variations on the idea that Gödel's theorem applies only to consistent systems, and human beings clearly do not think consistently.
I should note that I find this argument mysterious. If human minds are being inconsistent when they do advanced mathematics, then how do we manage to reach the same conclusions consistently? In any case, even the most committed Artificial Intelligence believers have mostly abandoned the idea that a program for an artificial intelligence can be written. Now they hope to create Darwinistic, self-programming systems that will organize an intelligent entity spontaneously. Good luck.
Gödel's theorem serves in popular culture as a symbol of the supposed irrationality of reality. As the authors note, the theorem tends to be dragged out these days to "hit people over the head" with. The authors are too polite to point out that this most subtle of logical arguments is often employed by persons who cannot make any logical argument at all. Nonetheless, it is clear that the theorem and the body of study it make possible are philosophically important, though people differ on just why. For me, the theorem is good evidence that the limits of language are not the limits of knowledge, or even of reason, broadly construed. This suggests that the world is objectively knowable. Surely this is a good thing.