Life, Intelligence, Choice and Will
By CosmicConservativeOr maybe I should just have titled this "Life, The Universe and Everything".
I just finished Kurzweil's "Singularity" book and find much in it to be fascinating, but I noticed something in his writing of it that makes me wonder how many people share what appear to be his key assumptions.
One of his major themes in the book, in fact the most profound theme, is that in the next few decades human life will be transcended by artificial intelligences. He does a solid job of supporting this assertion with facts, figures, quotations from major technology figures, projections based on historical progress, etc. His final assertion is that such intelligent machines will essentially be "human" in every meaningful way that we mean the word "human", primarily meaning that they will "think," will appreciate art, and will have all the subtle nuances of humans in social interactions. In fact he thinks they will vastly outstrip human capabilities even in these areas in a remarkably short time (a few decades).
One of the key points he tries to make is that once we make machines "intelligent" enough, we've essentially made them "alive", meaning he is equating "intelligence" with "life."
This seems a bit of an odd approach to take since everyone agrees that bacteria are "life" but nobody considers them "intelligent." So I'm pretty sure that something can be "alive" without being "intelligent", and there is no reason I can think of to assert that the reverse could also be true, that something could be "intelligent" but not be "alive."
In fact we can do a thought experiment right here to explore the relative differences between "alive" and "intelligent." And we can use the "Turing Test" to demonstrate that proving something to be of human intelligence doesn't necessarily prove it is "alive."
Imagine a massively parallel digital supercomputer with access to every bit of data on the internet. Imagine it to have a billion-line software program written from a reverse-engineering of the human brain, and for the purpose of this thought experiment, imagine it to be accurate to within .005% of an actual human brain.
Now, have someone ask it, and a human genius, a series of questions, some extemporaneous in nature to preclude the possibilitiy of the computer being able to look up the question and provide a rote, but reasonable, response. In this thought experiment, let us conclude that a series of these tests proves that the human genius and the computer are indistinguishable to the blind question asker. In other words, it passes not just the simple Turing test, but the "Genius Turing Test."
Now, that would prove, according to current accepted standards (the "Turing Test") that the computer is intelligent. But that's all it proves.
Now, imagine that we could run this experiment in such a way that we control all initial conditions and all variables. After running the experiment once, if we give the supercomputer exactly the same initial conditions and exactly the same questions in exactly the same order, then if the super computer responded exactly the same each time, we could conclude that the computer was undoubtedly exhibiting "intelligence" but that it was of a completely deterministic variety. Which would indicate that the supercomputer, for all its power, intellegence and complexity, was still essentially a "machine" which responded deterministically to its environment in exactly the same way a rock falls in a gravitational field.
Now, after having done that, we could introduce some sort of randomness to the computer's response. Let's say, for example, that we introduce a quantum device that returns a truly random result exactly comparable to flipping a coin. Now we put a rule in the computer's vastly complex decision-making tree that says "If any decision has two paths that are within .05% of each other in the decision making step, then instead of automatically taking the highest path, flip the coin and take one or the other based on the coin flip result."
This would change the results of the test, perhaps only slightly, but perhaps profoundly, depending on the question. Say, for example, the question is "how did the universe begin" and the computer's search revealed that 39.333% of its answers were "The Big Bang" and 39.332% were "God Created it", then the answer the computer would give to the question "How did the universe begin" would be "God made it" 50% of the time and "The Big Bang" 50% of the time. Obviously we can take this introduction of randomness to any level that we want to, but this is sufficient for this thought-experiment.
Now, doing this has demonstrated that an "intelligent" answer to a question can be different, and a system that answers differently based on purely random events would still pass the "Turing Test" each time, even if it provides different answers. We could call this random effect a manifestation of "choice". From an external perspective there would be no way to determine why one choice was made, but the effect of the choice would be clear in certain cases. But this sort of "choice" is based on pure randomness, so the question of whether the computer is "alive" is still not compelling enough to answer "yes."
And this is where "will" comes in. When we speak of "will" we are speaking of an entity making a choice based on something beyond deterministic or random processes. We understand "will" to be the manifestation of purpose. I make a choice because I want to affect the outcome, not because I am reacting to inputs.
When Kurzweil says that his intelligent machines will "transcend humans" he is assuming that they will manifest the property of "will".
Which brings us back to the bacteria. Anyone who has watched bacteria move around in a petri dish knows that they are alive. Their movements demonstrate that they are alive. And they demonstrate life by not being purely reactive to the environment. The movement is primarily predicated on stimuli, but not always. So this has caused some natural philosophers to speculate that even bacteria exhibit the property of "will," however rudimentary that may be. And so they assert that "will" is a purely biological phenomenon completely separate from "intelligence."
This is a crucial question for the science of AI. If we are creating fantastically sophisticated automata, that is different from creating life.
Kurzweil has a chapter on his critics, and most of the criticisms he spends any significant time on boil down to this fundamentally "bio-centric" argument, which essentially says that no matter how much a computer appears to be alive, no matter how "intelligent" it might be, it is still a machine, not a living thing.
I'm not so sure about this. Since we don't know what the source of "will" is, I think it is a bit anthropomorphic to assert that it comes from cellular activity of some sort.
But I'm not sure how we can ever know, since there is no way I know of to prove that will exists even for humans, and that has been debated for centuries already.
Comments (5)