1000 Words on Why I Don't Believe in Machine Consciousness
It is collective self-flattery for members of the computer science community<br>to argue that computers can be conscious. I will take the contrarian position<br>and argue that they cannot.
Arguments about machine intelligence hinge on questions of epistemology;<br>our ways of knowing what we know. The first and most basic argument of<br>this kind is known as the Turing Test. Alan Turing proposed that if a computer<br>was programmed in such a way that it could fool you, a human observer, into<br>believing that is was conscious, then it would be sentimental foolishness<br>to suggest that it wasn't conscious. It would be like claiming the Earth<br>was at the center of the universe; a desperate attempt to hold on to a vain<br>specialness.
I claim that there are different ways of knowing things. Consciousness<br>is the thing we share that we don't share objectively. We only experience<br>it subjectively, but that does not mean it does not exist.
How should we then approach the problem of deciding whether machines might<br>also experience consciousness? I believe the Turing test fails to help<br>us decide this question. In Turing's original setup, it is impossible to<br>tell whether the computer has become more human-like, or whether the human<br>has become more computer-like. All we are able to measure is their similarity.<br>This ambiguity makes artificial intelligence an idea that is not only groundless,<br>but damaging. If you observe humans using computer programs that are designated<br>to be "smart", you will see them make themselves stupid in order<br>to make the programs work. This can be observed in connection with Apple's<br>Newton or the program "Bob" from Microsoft.
What starts as an epistemological argument quickly turns into a practical<br>design argument. In the Turing test, we cannot tell whether people are<br>making themselves stupid in order to make computers seem to be smart. Therefore<br>the idea of machine intelligence makes it harder to design good machines.<br>When users treat a computer program as a dumb tool, they are more likely<br>to criticize a program that is not easy to use. When users grant autonomy<br>to a program, they are more likely to defer to it, and blame themselves.<br>This interrupts the feedback process that leads to improvements in design.<br>Any capability that might exist in a program that is designated "intelligent"<br>or "conscious" can also be presented in the context of a dumb<br>tool without ruining this vital feedback loop. The only measurable difference<br>between a smart program and a dumb tool is in the psychology of the human<br>user, and the dumb tool approach is the only option that leads to practical<br>improvements.
The above argument suggest that it is better for us to believe that computers<br>cannot be conscious, but what if they actually are? This is a different<br>kind of question, a question of ontology.
I argue that computers are not conscious because they cannot recognize each<br>other. In other words, if we sent a computer in a spaceship to an alien<br>planet and asked for a definitive analysis of whether there were computers<br>present on that planet, the computer would not be able to answer. There<br>are theoretical limits on one program's ability to fully analyze another<br>that make this so. People can nonetheless recognize and use computers,<br>so therefore people cannot be in the same ontological category as computers.
This is just another way of saying that without consciousness, the world<br>as we know it through our science need not be made of gross objects at all,<br>only fundamental particles. Complex processes such as computation can only<br>be found in nature through the filter of what Murray Gell-Mann has called<br>a "coarse-grained history" of the universe. For instance, one<br>has to be able to distinguish cars from air in order to measure "traffic".<br>Our most accurately confirmed scientific hypothesis, those of fundamental<br>physics, do not, however, acknowledge cars or other gross objects.
It is easy to claim that it is the state of a person's brain that notices<br>cars or computers, but that avoids the question of how the brain comes to<br>matter as a unit in the first place. If consciousness is associated with<br>a brain, why is it not also associated with a momentary correlation between<br>a brain and the arrangement of noodles on a plate of pasta being eaten by<br>the owner of the brain? If we wish to, we can find complex elements equivalent<br>to brains wherever we look. In a large enough city, the traffic patterns<br>could be interpreted as being exactly equivalent to a brain if we could<br>find the right translation table.
Even brains exist only by virtue conscious acknowledgment. The alternative<br>idea would be that the right kind of complex process gives rise to consciousness.<br>In that case there would be huge swarm of slightly different consciousness<br>around each person, corresponding to every combination of their brain, or<br>sections of it, with other objects in the universe. This idea violates<br>Occam's razor, the...