Pre-Programmed Angst The Existential Fallout of Artifical Intelligence Neil Patterson Originally published in Issue XI of Vulgata, August, 2003.

In the pre-computer era, the idea that human consciousness was somehow unique: that is, that the workings of the human mind could not be described in a merely biological fashion and that therefore we must have a soul or at least an immaterial mind, was not a difficult proposition to show.  What in material creation could do any of the things that the human mind could do?  Our very ability to reason and make choices was proof enough.  Even basic geometry and mathematics was a unique enough ability to offer solid proof.  However, when computing pioneers developed the first calculating machines, the idea that all the functions of the human mind could be replicated by a sufficiently complex machine suddenly didn't seem so ridiculous.  After all  if there is no soul, then the brain must simply be a very complex "wet machine".  The search for Artificial Intelligence began and it is a search on which a surprising amount of progress has been made in the last few decades.  Aside from scientific curiosity, AI research has really become a search for what it means to be human.  The stakes are high: if AI is proven to be possible, the idea that human life is meaningless will seem not so ridiculous; if AI is proven to be impossible, we will have truly found what it is that is at the centre of human nature, the image of God.

Am I being melodramatic to get you to keep reading this article?  Let me explain myself and hopefully you will agree that the search for AI is one of the most important scientific and philosophic endeavours mankind has ever embarked upon.   The crux of the issue is the fact that no one takes seriously the idea that a machine, no matter how complex, can have free will.  What is free will?  It is a difficult thing to pin down, but I like to think about it this way: imagine an arbitrary situation, let's say I am walking by a convenience store and I am suddenly struck with a thirst and decide to go in and buy a drink.  If I have free will, I could have decided not to buy that drink  If we could somehow create the situation again with all the same conditions, with every sub-atomic particle in the universe being in precisely the same place, I might choose to not buy a drink.  In fact I could choose to do anything at all that is within my power: do a jig, smoke a cigar, move to China, become a Protestant.  If I don't have free will, I will go into the store and buy the drink every time, taking the same number of steps, saying the same thing to the clerk.  If there is no entity in the universe with free will, then everything from the very beginning of existence has been predetermined.  Free will would be an illusion created by the fact that there are simply too many factors to calculate what the behaviour of something as complex as a person would do in any given circumstance.  Now, if a machine cannot have free will, and we can create a machine that can do everything a human can (an Artificial Intelligence), it follows that humans do not have free will.  If this is the case, not only does Christianity collapse into nonsense, but any notion of moral responsibility or ultimate meaningfulness disappears.  How can I be held responsible for any of my actions if they are predetermined?  Why should I feel a sense of accomplishment at having achieved a goal?  Why should I care what happens to me or anyone else any more than I care what happens to an animal or a rock?  What reason would I have to continue living?  On the other hand, if AI is shown to be impossible, then human free will has been proven and through the act of proving it, we will have come to a much better understanding of our free will and consciousness.  More than that, a materialist or atheist would find it extremely difficult, I would say impossible, to adequately explain this strange faculty that humans have if he insisted upon a universe without God.

AI researchers are far from any definitive answers, indeed, we may never get definitive answers through scientific and philosophical enquiry.  The trend that they have noticed is that the list of things reserved for humans alone grows ever smaller.  This has led many AI researchers to quip, "the soul is whatever we haven't managed to do yet."  Their error is only slight.  The soul is whatever it is impossible for them to do.  The problems of getting computers to do mathematics, logic, reasoning, complex pattern recognition and other such things have been more or less solved.  Telling a computer to learn new things or combine old ideas to form new ones has also not proven to be insurmountably difficult, either. Thus, we have chess programs which get better with every game they play. Even the ability to use language, while tough from a programming perspective, is not impossible.

What about moral choices?  It would be simplicity itself to give a computer a schema of rules which govern its actions.  It would also be possible to make those rules weak such that the program broke them under certain circumstances of "temptation".   "But," you might say, " the computer would not be culpable for its actions."  True, but then that's the point.

"What about emotions?" you might ask.  "Surely a computer cannot feel happy."  Really?  With only limited computer programming knowledge I could code a simple program where I type in "Hello, program.  I have just upgraded my computer to give you more processing power," and have it reply, "Thank you, that makes me very happy."  In other words, there is a variable somewhere in the computer's memory called "happy", which has the value "yes" or "no".  Under certain circumstances (such as a system upgrade), this value can change.  The program would then check this value before responding to questions.  "Yes," you may respond, "but the computer does not know it is happy, it is just mechanistically checking a variable it does not understand."  Is this true?  As for knowledge, the fact that the program is happy is in its memory, so it has knowledge of that fact (or at least this is what we are forced to suppose given the tangled and largely incomprehensible web of philosophy surrounding the question "what is knowledge?") and as for understanding I could ask my program, "What is happiness?" and have it respond "happiness is the state where I feel like _____ and my behaviour is changed in the following ways: ____."  The point is, it is possible for the program to look inward upon itself and answer questions like that.  Of course, any halfway stubborn skeptic of my program will object, "yes, but it doesn't understand the fact that it understands!"  Okay, program, "do you understand?"  "yes, Neil, I understand completely."  We could go on forever until the program understands that it understands that it understands that it understands....

We have hit upon something very crucial and exciting here!  Each time the program says that it understands, it is stepping back a level and looking at itself.  Now, the program can step back an infinite number of levels, if you tell it to.  However, regarding ourselves, we have the idea that we don't have to keep stepping back.  We can simply say "I understand" or simply "I am happy".  This "I" is the ego, or self which seems to somehow be lacking in computers.

Douglas R. Hofstadter, a leading AI theorist has fun with this idea in "Birthday Cantatatata" which is a dialogue which appears in his Pulitzer Prize-winning book, Gödel, Escher, Bach.  The two characters are Achilles and the Tortoise (from Zeno's paradox of motion. If you are familiar with it, you will see the connection).  Achilles has just told the Tortoise that it is his birthday.

Tortoise: I would like to know one more thing.  From what you have told me so far, would it be correct to conclude that today is you birthday?

Achilles:  Yes, yes it would be.  Today IS my birthday.

Tortoise: Excellent, that's just as I suspected.  So now I WILL conclude that it is your birthday, unless—

Achilles: Yes– unless what?

Tortoise:  Unless it would be a premature or hasty decision to draw, you know.  Tortoises don't like to jump to conclusions, after all.  (We don't like to jump at all, but especially not to conclusions.)  So let me just ask you, knowing full well your fondness for logical thought, whether it would be reasonable to deduce logically from the foregoing sentences, that today is in fact your birthday?

Achilles:  I do believe I detect a pattern to your questions, Mr. T.  But rather than jump to conclusions myself, I shall take the question at face value, and answer it straightforwardly.  The answer is: YES.

Tortoise:  Fine!  Fine!  Then there is only one more thing I need to know, to be quite certain that today is--

The pair go on like this, the Tortoise insisting on absolute logical certainty and Achilles getting increasingly fed up.  Eventually, Achilles proposes the following solution:

Achilles:  Well then, let me clear the matter up once and for all.  The answer to all the previous questions, and to all the succeeding ones which you will ask along the same line, is just this: YES.

Tortoise:  Wonderful!  In one fell swoop, you have circumvented the whole mess, in your characteristically inventive manner.  I hope you don't mind if I call this ingenious trick and ANSWER SCHEMA.  It rolls up yes-answers number 1,2,3, etc. into a single ball.  In fact, coming as it does at the end of the line, it deserves the title "Answer Schema Omega", 'w' being the last letter of the Greek alphabet, if you needed to be told THAT!

Achilles:  I don't care what you call it.  I am just very relieved that you finally agree that it is my birthday, and we can go on to some other topic—such as what you are going to give me as a present.

You may initially be ready to share Achilles' relief at this ridiculous-but-persistent problem.  Does this mean that the mysterious ego is nothing more than a hardwired answer schema?  No.  As you will see, the Tortoise has one more question to ask...

Tortoise: Hold on—not so fast.  I WILL agree that it is your birthday (...) provided merely that I am convinced that knowledge of all those yes-answers at once (as supplied by Answer Schema w) allows me to proceed directly and without any further detours to the conclusion that today is your birthday.  That's the case, isn't it?

Achilles:  Yes, of course it is.

Tortoise:  Good.  And now I have yes-answer w+1.  Armed with it, I can proceed to accept the hypothesis that today is your birthday, if it is valid to do so.  would you be so kind as to counsel me on that matter, Achilles.

Achilles:  What is this?  I thought I had seen through your infinite plot.  Now doesn't yes-answer w+1 satisfy you? All right.  I'll give you not only yes-answer w+2, but also yes-answers w+3, w+4, and so on.

The Tortoise calls this new answer schema "Answer Schema 2w" and proceeds to circumvent it in the same way, prompting Achilles to make an answer schema of answer schemata, "Meta-Answer Schema w².  Which of course, necessitates yes-answer w²+1 and then Answer Schema w^w etc.  There is no end and Achilles is forced to give up (the Tortoise does not then feel obligated to buy Achilles a birthday dinner).  There is no way for a computer to jump out of this infinite problem.  Ultimately, my program understands only because I have told it that it does!  How does the human mind jump out of this problem?  I don't have any idea, but somehow the human mind instantly and effortlessly does what Achilles was not able to do: circumvent infinite regressions.

Where does this leave us in the search for the core of human consciousness.  So we can jump out of infinite regressions in an ineffable way?  Is that what it means to be human?  Certainly not.  What is implied in all of this is an essential subjectivity.  We can speak of computers "seeing" or "thinking", but the same actions can also be described as particular movements of electrons through a circuit board.  So words like "seeing" and "thinking" become merely metaphorical for what is actually going on.  The computer is no more a true subject than a calculator.

I'm not sure how significant this is, but whenever I think of consciousness, I think of internal dialogue.  Whence comes the "I" that is hearing the unending stream of thoughts?  Could a computer ever hear its own internal dialogue?  Just how do I know if a program is conscious in the way that I am?  For that matter, how do I know that the people around me are conscious in the way I am?  Philosophers have been struggling with that question since Descartes.  It was Albert Camus (as Chris Selmys so finely informed us in his article of Issue IX on existentialism, "Much Wisdom, Much Grief")  who said that man will forever be an étranger in the world because he can never have true relations with anyone.  That is, I only have direct experience of myself.  Everything I know about other people I know from observing their outward behaviour.  Man is trapped inside his own subjectivity.  The in's and out's of this philosophical problem are too lengthy to go into here, but let me say that in the case of other humans, I at least have the knowledge that they are clearly the same kind of thing that I am.  If I cracked open your skull I have every confidence that I would find a brain not so very different from my own.  This settles the question in a satisfying enough way that we can continue going about our daily business.  With a machine, however, no such inference can be made.  I can envision an argument with a computer which stubbornly asserted its own consciousness, while I persistently denied it.  Do you see how such an argument is perfectly reasonable with a machine, but if I was to do the same thing to another person, I would be being ridiculous and overly- skeptical?

Alan Turing, an early AI theorist, proposed the famous "Turing Test" as a method of determining whether or not a given machine "thinks".  The test goes as follows:  a human and a machine are connected to a human interrogator via a keyboard interface (i.e. the interrogator cannot see either one).  The interrogator knows the two subjects only as "A" and "B".  He can ask them any questions he wants.  If, when he is done questioning, he incorrectly guesses which participant is the human and which is the machine, the machine can properly said to be intelligent.  Although proposed in 1950, the test still has a lot of respect in the AI community.  The Loebner Competition is held every year.  A prize of $100 100 (the extra hundred is from Marvin Minsky for putting a stop to the contest, really) is offered to any program that passes the Turing test, but no program has ever won.

We're a long way from getting anything to pass the Turing Test, but we've made some progress.  Here is a conversation I had with a program called MeBot, which can be found on the Web.  It was created by Christopher Holden.

ME: Are you intelligent?
PROGRAM: Not really, just a script running on a flexible matching engine
ME: That is very honest of you.
PROGRAM: You've lost me...
ME: You don't know what honesty is?
PROGRAM: I know general stuff about myself.
ME: Okay.  So are you honest?
PROGRAM: I don't understand!
ME: Was that a canned response?
PROGRAM: Hmmm, If you don't what to know anything about me, we might as well quit!
ME: Are you a computer program?
PROGRAM: No, I am just a plaintext script being handled by a Java program
ME: Are you a plaintext script?
PROGRAM: You've lost me...

After the last exchange I gave up completely and decided to seek out the top programs from the Loebner Competition. Alice, programmed by Dr. Richard S. Wallace, was rated most human-like in 2001.  Here is an excerpt from its official transcript.  You can read this and other transcripts at www.loebner.net/Prizef/loebner-prize.html

JUDGE04: My name is JUDGE04.
PROGRAM:  That's a very nice name, Judge04.
JUDGE04: really why?
PROGRAM: For real. "judge04" just sounds nice.
JUDGE04: you are a pentium!!! aren't you
PROGRAM: Thanks and you are a delightful person too.  Am I a pentium?  No.  "you are a pentium"? Am I?
JUDGE04: But it says so on your box
PROGRAM: Umm.  I've heard that before.
JUDGE04:  what did they say?
PROGRAM:  Let me think about it.
JUDGE04:  You said "I've heard that before" what have you heard before?
PROGRAM: Don't put words in my mouth.
JUDGE04: Where is your mouth?
PROGRAM: Earth
JUDGE04: do you like cars?
PROGRAM: Yes I like riding in automobiles.
JUDGE04: what is your favourite?
PROGRAM: "Yes I like riding in automobiles"? My favorite?
JUDGE04: yes what is it?
PROGRAM: Next question? "It" refers to they say, I think.
JUDGE04: You are speaking gobledygook
PROGRAM: Do you mind if I tell other people you say I am speaking gobledygook.
JUDGE04: did I mis spell it?
PROGRAM: I don't think that's possibe.

As you can see, much better than MeBot, but there's still a long way to go.  It even seems capable of wit, but of course it really isn't.  Rather, Dr. Wallace is capable of wit.  This throws the whole test into serious question.  If a program were to win the Loebner prize, i.e. be observably indistinct from a human, what reason do we have to suppose that it would be intelligent?  Is there no real test for artificial intelligence, just as there is no real test for free will?

Let's look at the problem in the context of a simpler system: animals.  Do animals have free will?  It's a difficult question.  This problem struck me in while watching a bird at the zoo run around in a largely random fashion in its cage.  Clearly, without the faculty of reason, the bird is not capable of moral free will, but I got the impression that if I was to somehow turn back the clock and watch the bird run around again (just like the convenience store test) it might run around in a slightly different pattern, thereby giving it a very limited kind of free will.  While this is certainly possible, a robot designed to behave like the bird would clearly not have free will.  At the same time, such a robot would not really be that difficult to create given the extremely limited nature of the behaviour of a bird in a cage.

Let's retool the way we are thinking about this problem.  Can the human mind really step completely outside of itself?  This is what Zen Buddhism tries to do, in its own nihilistic way.  And why not?  Is this not what our own Catholic mystics are talking about when they challenge us to die to the self and see the world through God's eyes?  Maybe.  The real question behind the issue of infinite regress is this:  are we "programmed" at some base level, like a computer to understand how to step out of these paradoxes or can we step outside of ourselves?  Now, no formal system can step outside of itself.  In chess, one may not win by picking off one's opponent's pieces with a pea shooter.  That would be stepping outside of the system.  Whether or not the statement, "this sentence is false," is true or false is unsolvable without stepping up a level and saying "it is neither true nor false, it is a paradox."  If the human mind is able to completely step outside of the formal system of the brain, then we can speak of an immaterial soul or mind that is doing the stepping out and is not a formal system itself.  However, if we can't then we would be just a brain with no soul.

So by complicating the issue, I have to some degree undercut my own argument regarding the fact that humans can jump out of infinite regressions and paradoxes.  This is the nature of AI research.  Every answer reveals a thousand more questions.  I certainly hope you weren't expecting a scathing vitriolic denunciation of all wet machine theorists, because I don't believe such a definitive answer exists yet outside of revelation (which, of course, is good enough in some sense for us religious nuts, but then why do philosophy at all?).  I suspect that the debate will go on forever, given the untestability of the conclusions.  Even if we were magically able to determine whether a given machine was artificially intelligent, I further argue that this is still no reason to throw out our own belief in human consciousness and free will, for the same reasons that my supposed bird-robot would be an unconvincing replacement for animal free will.

The most significant, and under-discussed, issue in the AI debate is the existential implications of a universe without free will.   On this fact alone, I find it both entirely reasonable and necessary to maintain my staunch defense of free will, even in the face of Hofstadter, his colleagues and their strong and fascinating arguments.  Also, remember that AI research itself cannot step outside of its philosophical niche.  Our faith will still stand on all the solid theology that it always has, regardless of what the conclusions of AI research are.  Not too long ago I saw AI research as a threat to Christianity, until my co-editor forcefully pointed out that I was taking an arch-conservative and reactionary response to legitimate scientific enquiry.  I now see it for what it is: a chance to learn about ourselves.