Darwin's Dangerous Idea (85 page)

1968).

444 THE EMPEROR'S NEW MIND, AND OTHER FABLES

The Phantom Quantum-Gravity Computer
445

3. THE PHANTOM QUANTUM-GRAVITY COMPUTER:

Moreover, in the wake of the commentary his book provoked, Penrose LESSONS FROM LAPLAND

now grants that heuristic programs are algorithms as well, and acknowledges that, if he is to find an argument against AI, he has to concede their tremendous power to track the truths of arithmetic and everything else, if not
I am a strong believer in the power of natural selection. But I do not see
perfectly, then at least impressively. He offers a further point of clarification:
how natural selection, in itself, can evolve algorithms which could have
any computer that operates by indulging in interactions with an external
the kind of conscious judgements of the
validity
of other algorithms that
environment is an algorithmic computer
provided the external environment is
we seem to have.

itself entirely algorithmic.
(If skyhooks grew like toadstools—or, more to the

—ROGER PENROSE 1989, p. 414

point, like oracles perched on toadstools—and a computer was helped along by its occasional communication with these skyhooks, then what it did would

/
don't think the brain came in the Darwinian manner. In fact, it is
be no algorithm.)

disprovable. Simple mechanisms can't yield the brain. I think the basic
Now, with all this useful clarification in place, what does Penrose main-elements of the universe are simple. Life force is a primitive element of tain? In May 1993, I spent a week with Penrose and some Swedish physicists
the universe and it obeys certain laws of action. These laws are not
and other scientists discussing our different views about these matters, at a
simple and not mechanical

workshop
in
Abisko, a tundra-research station well
north of the Arctic
Circle

—K

2

in Sweden. Perhaps the midnight sun helped as much as our Swedish hosts to URT GODEL

illuminate the path, but, in any event, I think we both came away When Penrose insists that the brain is no Turing machine, it is important to enlightened. Penrose proposes a revolution in physics, centered on a new—

understand what he is
not
saying. He is not making the obvious (and and still unformulated—theory of "quantum gravity," which he hopes will obviously irrelevant) claim that the brain is not well modeled by Turing's explain how the human brain transcends the limitations of algorithms. Does original thought-device: a smallish gadget sitting astride a paper tape, ex-Penrose envisage the human brain, with its special quantum-physics powers, amining one square of the tape at a time. Nobody ever thought otherwise. He to be a skyhook or a crane? That was the question I went to Sweden to answer, and the answer I came back with is this: He has definitely been is also not merely saying that the brain is not a serial computer, a "von looking for a skyhook. I think he'd settle for a new crane—but I doubt that Neumann machine,'' but, rather, a massively parallel computer. And he is not he's found one.

just saying that the brain makes use of randomness or pseudo-randomness in Descartes and Locke, and more recently Edgar Allan Poe, Kurt Godel, and running its algorithms. He sees—though some others have not—that J. R. Lucas, thought that the alternative to a "mechanical" mind would be an algorithms availing themselves of large doses of randomness are still
immaterial
mind, or a soul, to speak with tradition. Hubert Dreyfus and John algorithms within the purview of Artificial Intelligence, and still fall under Searle, more recent skeptics about AI, have shunned such dualism and opined the limitations Godel's Theorem places on all Turing machines, of whatever that the mind is indeed just the brain, but the brain is not any
ordinary
size and shape.3

computer; it has "causal powers" (Searle 1985) that go beyond the running of any algorithms. Neither Dreyfus nor Searle has been very forthcoming about what special powers these might be, or which of the physical sciences might be the right one to give an account of them, but others have wondered 2. A remark made in 1971, quoted in Wang 1993, p. 133. See also Wang 1974, p. 326: whether physics might hold the key. To many of them, Penrose appears to be

"Godel believes that mechanism in biology is a prejudice of our time which will be a knight in shining armor.

disproved. In this case, one disproval, in Godel's opinion, will consist in a mathematical Quantum physics to the rescue! Several different proposals have been theorem to the effect that the formation within geological times of a human body by the laws of physics ( or any other laws of a similar nature ), starting from a random distribu-advanced over the years about how quantum effects might be harnessed to tion of the elementary particles and the field, is as unlikely as the separation by chance give the brain special powers beyond those of any ordinary computer. J. R.

of the atmosphere into its components."

3. Someone who doesn't realize this is Gerald Edelman, whose "neural Darwinism"

everybody in AI knows, that although you may not have "Absolute Ignorance" (as Mac-simulations are both parallel and heavily stochastic (involving randomness), a fact he Kenzie anonymously put it, back in chapter 3, p 65), you still don't have to understand often cites, mistakenly, as evidence that his models are not algorithms, and that he himself what you are making in order to make it.

is not engaged in "strong AI" ( e.g., Edelman 1992 ). He is; his protestations to the contrary betray an elementary misunderstanding of computers, but that just goes to show, as 446 THE EMPEROR'S NEW MIND, AND OTHER FABLES

The Phantom Quantum-Gravity Computer 447

Lucas (1970) yearned to drag quantum physics into this arena, but he thought
Final Theory
(Weinberg, you will recall from chapter 3, gave two cheers for that the indeterminacy gaps of quantum physics would permit a Cartesian reductionism), Penrose mused as follows:

spirit to intercede, twiddling the neurons, in effect, to get some extra mind-power out of the brain, a doctrine that has also been energetically defended In my view, if there is to be a Final Theory, it could only be a scheme of by Sir John Eccles, the Nobel-laureate neurophysiologist who has a very different nature. Rather than being a physical theory in the ordinary scandalized his colleagues for years with his unabashed dualism (Eccles sense, it would have to be something more like a principle—a mathemat-1953, Popper and Eccles 1977). This is not the time and place for me to ical principle whose implementation might itself involve nonmechanical review the reasons for dismissing this dualism—the times and places are subtlety (and perhaps even creativity). [Penrose 1993, p. 82.]

Dennett 1991a, 1993d—since Penrose shuns dualism as vigorously as anybody else in the materialist camp. What is refreshing about his attack on So it is not surprising that Penrose has expressed grave skepticism about AI, in fact, is his insistence that he hopes to replace it with something that Darwinism. And the grounds he gives are familiar: he can't imagine how would still be a physical science of the mind, not some unexplorable mystery

"natural selection of algorithms" could do all that good work: that takes place in the never-never-land of dualism.

[T]here are serious difficulties with the picture whereby algorithms are Without abandoning the physical sphere, we might get some strange new supposed to improve themselves in this way. It would certainly not work powers out of subatomic particles, according to recent speculations about for normal Turing machine specifications, since a 'mutation' would almost

"quantum computers" (Deutsch 1985). Such a quantum computer would take certainly render the machine totally useless instead of altering it only advantage (it is claimed) of the "superposition of eigenstates" prior to the slightly. [Penrose 1990, p. 654.]

"collapse of the wave packet" in order to check out Vast (yes, Vast) search spaces in ordinary amounts of time. By being a sort of supermassively parallel Most mutations, Penrose sees, are either invisible to selection or fatal; only a computer, it could do Vastly many things "at once," and this could render very few improve things. That is true, but it is just as true of the evolutionary feasible whole classes of algorithms that otherwise were unfeasible—such as processes that produced the mandibles of crabs as it is of those that produced the algorithm for perfect chess. This is
not
what Penrose is seeking, however, the mental states of mathematicians. Penrose's conviction that there are these for such computers, even if they are possible, would still be Turing machines,

"serious difficulties" is undercut, as Poe's conviction was, by the brute and hence capable of computing only the officially computable functions—

historical fact that genetic algorithms and their kin are daily Overcoming the algorithms (Penrose 1989, p. 402 ). They would hence fall under the these fearsome odds and improving themselves by, well, leaps and bounds limitations discovered by Godel. Penrose is holding out for a phenomenon (on the geological time scale).

that is truly
noncomputable,
not just impractical to compute.

If our brains
were
equipped with algorithms, Penrose argues, natural Present-day physics (including present-day quantum physics) is
all
com-selection would have to have designed those algorithms, but: putable, Penrose acknowledges, but he thinks that we might have to revolutionize physics, incorporating an explicitly noncomputable theory of The 'robust' specifications are the
ideas
that underlie the algorithms. But

"quantum gravity." Why does he think such a theory (which neither he nor ideas are things that, as far as we know, need conscious minds for their anyone else has yet formulated ) would have to be noncomputable? Because manifestations. [Penrose 1989, p. 415]

otherwise AI is possible, and he thinks he has already shown, via his argument from Godel's Theorem, that AI is not possible. That's all. Penrose In other words, the designing process would have to appreciate, somehow, candidly admits that none of his reasons for believing in the noncomput-the rationale of those algorithms it was designing, and doesn't that take a ability of quantum-gravity theory are drawn from quantum physics itself; the conscious mind? Could there be reasons recognized without some conscious mind's recognizing them? Yes, says Darwin, there could be. Natural
only reason
he has for thinking that a theory of quantum gravity would be selection is the
blind
watchmaker, the
unconscious
watchmaker, but still a noncomputable is that otherwise AI would be possible after all. In other discoverer of forced moves and other Good Tricks. This is not as incon-words, Penrose has a hunch that someday we're going to find a skyhook. This ceivable as many have taken it to be.

is the hunch of a brilliant scientist, but he himself admits that it is only a hunch.

To my way of thinking, there is still something mysterious about evolution, In a review of the physicist Steven Weinberg's recent book,
Dreams of a
with its apparent 'groping' towards some future purpose. Things at least 448 THE EMPEROR'S NEW MIND, AND OTHER FABLES

The Phantom Quantum-Gravity Computer 449

seem
to organize themselves somewhat better than they 'ought' to, just on reliable shortcuts to verdicts about projects. If someone were to go to the the basis of blind-chance evolution and natural selection. It may well be Swedish government with a plan to build a perpetual-motion machine (at that such appearances are quite deceptive. There seems to be something government expense), Hansson would unhesitating testify, as a physicist, about die way that die laws of physics work, which allows natural selection that this would be—would
have to be
—a waste of government money. It to be a much more effective process than it would be with just arbitrary could not succeed, because physics has proven that a perpetual machine is laws. [Penrose 1989, p. 416.]

flat impossible. Did Penrose think that he had offered a similar
sort
of proof?

If some AI entrepreneur were to go to the government asking for money to There could not be a clearer, more heartfelt expression of the hope for build a mathematical-truth-detecting machine, would Penrose be similarly skyhooks than this. And though we cannot yet rule out "in principle" the willing to testify that such money would be wasted?

existence of a quantum-gravity skyhook, Penrose has not yet given us any To make the question more specific, consider some rather special varieties reason to believe in one. If his theory of quantum gravity were already a of mathematical truth. It is well known that there can be no all-purpose reality, it could well turn out to be a crane, but he hasn't got that far yet, and I program that can examine any other program and tell whether or not it has doubt that he ever will. At least he's trying, however. He wants his theory to an infinite loop in it, and hence will not stop if started. This is known as the provide a unified, scientific picture of how the mind works, not an excuse for Halting Problem, and there is a Gödel-style proof that it is insoluble. ( This declaring the mind to be an impenetrable Ultimate Source of Meaning. My is one of the theorems Turing alluded to in his 1946 comment quoted at the own opinion is that the path he is now exploring—in particular, the possible beginning of the chapter.) No program that is itself guaranteed to terminate quantum effects occurring in the microtubules of the cytoskeleton of can tell of every (finite) program whether or not it will terminate. But it neurons, an idea enthusiastically promoted in Abisko by Stuart Hamer-off—

might still be handy—worth some serious money—to have a program is a nonstarter, but that is not a topic for this occasion. (I can't resist raising around that was very, very good (if not perfect) at this task. Another class of one question for Penrose to ponder: if the magnificent quantum property interesting problems are known as Diophantine Equations, and it is known lurks in the microtubules, does that mean that cockroaches have that there is no algorithm guaranteed to solve all such equations. If our lives noncomputable minds, too? They have the same kind of microtubules we depended on it, should we spend a nickel on a program for solving Diophan-have.)