Sep. 22, 2000 THE NUMBER OF THE BEAST

There is a burgeoning school of scientific/mathematical/philosophical thought propounding a thesis so preposterous on its face that it is seldom put as baldly as I am about to put it. Yet I believe this thesis is equivalent to the following:

If you could write down a sufficiently long list of ones and zeroes, and have them in the right order, that list could not only describe (say) a particular individual and his environment and his lifetime experiences, but would constitute that person and his experiences.

The "strong AI" people or "uploaders" believe, in effect, that a person is basically just code. Not only are you "just" a computer; you are not even anything material in essence, not even a computer, but basically only a computer program-just a set of numbers! If the set of numbers representing your personality and memories could be copied or transplanted into another vehicle, organic or not, then you would survive or/and be duplicated.

Furthermore, since any set of numbers could be said to "exist" even if not written down, one extreme viewpoint must be the notion that everything logically possible actually exists-and always has and always will. This goes beyond even the Multiverse interpretation of quantum theory.

After decades of discussion and many shelves sagging with books and learned journals, there is no agreement on this issue, nor on its consequences, nor even on whether the issue is important. My own opinion is that it is important, and the uploaders are probably wrong. It isn't easy to get a clear understanding of the arguments, but let's make a start and work into it.

An ancient partial precursor of the uploader school was that of the philosopher Plato, who held that, beyond or above the material world, there is another world of forms and ideals. The latter is exemplified most obviously in the world of mathematics.

For example, there are many sets or collections of physical objects containing three items. But nowhere in the physical world will you find the number three, although you will find words and symbols for it. Three itself (or "threeness" if you will) is an abstraction existing only in Plato's world of ideals.

Yet it really and truly does exist in an absolute sense, according to the Platonists. It is not just a human invention or a way of speaking. They make their point, for example, by reminding us that Pythagoras discovered his theorem about right triangles; he did not invent it. It was there all the time, even before humans existed, and could have been discovered by someone else. It is an eternal truth, not just a construct.

This example has taken a wry twist in the last century or so, with the discovery that, in fact, Euclidean geometry and the Pythgorean theorem do not hold exactly in our physical world. The prevailing opinion is that the geometry of our physical universe is Einsteinian or Minkowskian or curved, not Euclidean or flat. So the philosophy moves to another level. In what sense is something "true" if it has no counterpart in physical reality? As usual, there is no agreement, but we can certainly allow at least the possibility that imaginary philosophical or mathematical structures can have some kind or degree of validity.

Even more fundamental, perhaps, is the status of logic itself, since logic underlies all other sciences and cognitive efforts. How solid are the foundations of Aristotelian logic? Some writers-Bart Kosko is one who is well known-think that nature is always "fuzzy" and no physical quantity is ever perfectly defined, so that the true/false dichotomy at the root of Aristotelian logic is only an approximation or abstraction, and can be misleading. We bypass this for the moment, noting only that Aristotelean logic and probability theory, together, may be able to accomplish everything that "fuzzy logic" can do.

Bekenstein, Tipler, Moravec, Kurzweil--Soul of the Machine

Frank Tipler, Hans Moravec, and Ray Kurzweil are a few of the distinguished writers who believe that "you" could be "uploaded" into a computer (in principle, at some future time when the requisite technology is available), and that such duplication would represent survival-and indeed much more than survival, since after that you might be enormously improved as well as immortal.

Computer science finds an ally in quantum theory in the form of the Bekenstein Bound-a very, very large number which represents the maximum number of possible states of a human-sized brain, with the profound implication that this number is the maximum number of possible experiences for a brain of human size. If you could ever run through that repertory, then after that anything would be repetition, unless your brain could grow in size. At the moment, however, we are concerned more with the basic philosophy and concept of self and personhood.

In some religions we have the notion of a "soul" that is distinct from the corporeal person, and might survive death of the body-might in fact be eternal. (In most cases it is conceived as being just semi-infinite in time, extending from conception or birth into the endless future, but not existing prior to conception or birth.) Is there a partial parallel in uploading?

Yes, both this religious notion and Plato's "ideal" have something in common with the uploading thesis. If you are basically just a computer program-a set of numbers-then in some sense you have always existed and always will exist (along with previous and subsequent versions of you), whether or not you are embodied in a physical instantiation. To be and potentially to be may overlap. If you believe the number three "exists" without regard to anything in the physical world, then you might well believe that you exist-now and always-regardless of any untidy structures and their behavior in the physical world. Has science found soul?

I doubt it, for reasons we can perhaps understand best by looking at some of the consequences of Alan Turing's ideas.

Turing Tape and Turing Tome

One of the giants of computer theory, Turing devised the Turing Tape (TT) as an exemplar of the Universal Sequential Digital Computer. The details are not important, and the Tape would be hopelessly slow and cumbersome in practice, but in principle it can do anything that any sequential digital computer could do. It is just a very long paper tape, divided into squares, with auxiliary equipment. Each square can have a zero, a one, or nothing written on it. The device moves the tape back and forth, one square at a time, and may write or erase ones and zeroes, according to a program (algorithm) of instructions. This program is itself represented by some of the original marks on the tape.

Initially the tape holds a data store including both the program and other items of memory. Once it is started, everything is automatic, and the computer keeps computing until it produces the desired result-a number or set of numbers. If it is correctly programmed yet fails to reach the result-either because that would take infinite time, or because the task is impossible for some other reason--it just keeps running. It might run in circles after a while, or it might keep generating larger segments of infinity, say if asked to compute the decimal expansion of pi.

Now I hold that a TT cannot constitute a person, or a person's life, for reasons including the following.

First, there is good reason to believe that the human brain, in order to have experiences or qualia or feelings, must have more than one thing happening simultaneously in its physiology, not just a point event. (I have suggested that the "self circuit" may be a kind of standing wave in the brain.) But in a sequential digital computer, attempting to simulate a person, we have only a sequence of numbers being generated, one at a time, each number typically corresponding to only a tiny fraction of the quantum description of a brain or to the anatomy/physiology of the brain.

The brain does have certain aspects or functions that remind us of computers, and computers sometimes remind us of brains, but there are sharp differences between the brain and any existing computer, and especially any sequential computer. For one thing, the brain uses parallel processing, not sequential processing. It also uses analog computation, instead of or in addition to digital computation.

"Stop!" complain the uploaders. "It doesn't matter about parallel processing," they say. A TT is sequential, but it can nevertheless do anything that any digital computer-including a parallel processor-can do. Any little old TT is equivalent-except for speed-to any imaginable digital computer or network of them. The redoubtable Alan proved it.

Well, yes and no. Yes, the TT can (eventually) produce any number or set of numbers that any other digital computer, sequential or parallel, might produce. But if you claim that this is all that is necessary-to eventually produce the right sets of numbers-then you are making an extraordinary assumption which most people will find absurd. You are assuming that time doesn't matter.

Time is profoundly mysterious, and likely to remain so for the foreseeable future. It is also interwoven with most of our deepest questions. Hence anyone who claims to have final answers to these questions is egregiously reckless. Let's look now at some of the ways in which time enters the mind/computer controversy.

The uploaders claim that a person could in principle be simulated or emulated (perfectly simulated) in a computer, so that the running computer would constitute a living person, and in fact that particular living person, with all his thoughts and feelings. (I neglect here such details as the necessity of having the computer also simulate the whole environment.)

Yet what is actually happening while the TT computer is running? The tape is moving back and forth in a jerky motion, and marks are being made on its squares or erased. Think of the consequences.

First, every imprint or erasure will usually be only a portion of an aimed-at number, an intermediate step in a calculation. And even that aimed-at number will be only a portion of the aimed-at set of numbers required (for example) to specify any vector, such as the velocity of a particular solute molecule in a particular neuron at a particular time. Hence the uploader is assuming, in effect, that there is no subjective change at all in the simulated person's mind until-momentarily, and if that ever happens-all of the requisite sets of numbers on the tape suffice for a completely designated mental state.

To further complicate the question of correlation between states of the computer and mental states of the simulated person, we can ask what happens if the computer stops for a while. Physically, it does stop, or at least hesitate, between displacements of the tape, or/and while marks are being made or removed; so when we ask what happens if there is a longer pause, that isn't much different.

In other words, when does the simulated person have his subjective experiences? Surely not at a mathematical instant of time (if that has any physical meaning), since most of the time no change is being made to the mental configuration set of numbers-almost all of the time, nothing is happening except intermediate calculations in the computer, or a slow writing or erasing of marks on the squares. If there is a persuasive answer to such questions, lending plausibility to the mind-in-the-computer, I haven't seen it.

It is sometimes claimed that similar problems exist for ordinary brain biology; I disagree, but bypass that for the moment.

The Turing Tome

Let's try another angle of attack. The reader may have noticed that the uploaders claim, in effect, that a description of a thing is as good as the thing itself, or even the same as the original. Isomorphism is everything (only correspondences matter), and the map is the same as the territory. Symbolic blood will sustain symbolic life, and the symbolized person will be real, with precisely the same joys and sorrows as the original. ("Do I not bleed? No. But I can bleed to death, all the same.")

This is not quite as preposterous as it may at first seem to the novice. After all, much of what goes on in our heads is "merely" symbolic.

A visual perception, for example, is the result of a very complicated process in which light is reflected or otherwise transmitted from an external object, enters the eye, is changed into a chemical/electrical signal, transmitted to the appropriate part of the brain, processed, and finally interpreted subjectively in some cognitive context.

Also, think of a computer controlling an oil refinery. Even though its internal action is only a shuffling of symbols, nevertheless it is exchanging information with the outside world, and achieving a degree of control over external physical reality.

Further, some smart people believe we could write a code or language capable of being deciphered-without any Rosetta Stone-by any intelligent species. Details would be space consuming here, but I believe they are correct. In other words, the "mere" relationships between symbols could prove capable only of a single, unique interpretation. This would be a heavy blow against those-like John Searle-who claim that any computer, "merely" processing symbols subject to interpretation, by itself can only display syntax, never semantics.

But let's get to the Turing Tome. This is just an enormous book--or planet covered with books, whatever--containing a set of numbers adequate to represent a complete individual and at least a fair portion of his history and environment. Since we have isomorphism--a one-to-one correspondence between the states representing a possible physical person and the numbers in the Tome--we conclude (taking the Uploaders' view) that the Tome is the person, or at least an instantiation of the person. A paper person is just as alive and valid as a meat person.

When Uploaders are asked to respond to this, sometimes they are worried by the problem of statics vs. dynamics and say, well, the book can't just lie there; there has to be some activity. Maybe someone has to turn the pages before the embedded person feels anything? But that response is faulty as well as feeble. Our isomorphism includes time. If you can take the giant step of allowing symbols instead of physical matter, then you can hardly balk at allowing relations between symbols instead of physical time.

Upshot?

As usual, my only firm conclusion (with apologies to Gödel's ghost) is that we can as yet draw no firm conclusion. But I strongly suspect that the meat of the matter of personal existence is in our meat, not in Plato's world of abstractions.

--R.C.W. Ettinger