Artificial Intelligence and Rationalism

You may not be aware of this, but there is actually an organization that is working to protect mankind from a self-aware artificial intelligence (AI) that will be infinitely more intelligent than we are, and thus a huge threat to our existence. It is called the Singularity Institute and they are taking donations. When I first heard about this effort, I wondered if it was real, and when I then heard that their main group of benefactors come from the rationalist community, I was flabbergasted. Artificial Intelligence is a field where huge leaps of innovation are sure to be made in the future, but the highly specific worry of a self-improving AI that can potentially destroy our civilization is straight out of science fiction. The fact that otherwise rational and intelligent people are putting their money into this cause is more than a little disheartening.

It’s a great topic to muse about, though. Nearly all of us have read stories or seen movies about robots and computers that think thousands of steps ahead of us, which end up making decisions for us and controlling our actions to their own benefit. We all have had our imaginations seeded with this idea and so it is not easy to stop ourselves and think, “but then again, this is all speculation.” The line of thought usually goes as follows: if we can build a self-improving AI, then it is then possible for the AI to continually improve. Given enough processing power and memory, an AI could potentially raise itself up to the same level of intelligence that we have and then go even further. Once the AI is so far beyond our intelligence, it would then be able to manipulate us in ways we cannot even predict. This is the fear. This is the fiction.

From a rational perspective, there are so many things wrong with this idea, that it makes it almost too easy to pick apart. The most basic of these being the fact that we have never witnessed a greater intelligence, therefore we have no evidence that one’s existence is possible. Not only is the possibility of greater intelligence pure speculation, but the capabilities, intentions, and logical conclusions that a super-intelligent AI might reach are also pure speculation. I called it a specific worry because, given the nature of reality to manifest itself in particularly novel ways, we somehow think that we’ll give birth to Skynet by accident. Allow me to call it the “god program.” Its powers are mythical; they demand respect in so far as we can imagine them being beyond our grasp. From an investment standpoint, it seems like a poor way to spend time, money and effort.

I’m not saying that it is impossible for a higher level of intelligence to exist, I’m only saying that there is no precedent for it. Anyone familiar with the chaos that pervades the happenstances within reality would find it a far better cause to, say, come up with a survival plan in the case that Yellowstone blows its top, or a giant meteor crashes into the Earth. You know, let’s worry about things that have happened many times before and are almost certain to happen again. The risk of a malicious AI that we cannot control can be filed away next to the Extraterrestrial Diplomacy Initiative and the time travel survival kit.

Let’s play the game though. Let’s assume that computer development and brain research will continue on its path. We will become more capable with each passing year until, eventually, we can both model every neuron within our brains and conduct them, like a symphony, within a computer model and thus create an intelligence equal to ours. Then, given the predicted improvements in computer processing power, this artificial brain will race ahead, leaving our sad monkey brains in the evolutionary dust. It will find the best ways to learn and it will mold itself into a beast of a thinking machine. Of course, even laying out this path brings to mind a hundred logical gaps that I’m ignoring. For one example: much of human progress is made through experimentation within reality, but a computer is only able to work with known models that are programmed into it. Without inputs and outputs that would allow the computer to interact with the world on the same level of intricacy that we do, it would have no way of learning any more than its own pre-programmed environment.

For an intelligence to outwit us, it must first know us. And not just one of us; as many profiles as can be assimilated. It must be able to observe us on a macro scale, and then, here’s the crux, experiment on us. It simply does not follow that an AI can self-improve within a vacuum. It’s predicted that computers will have the same processing power as a human brain by the year 2029, but power is useless without discipline and focus. While all the great minds are busy writing papers about what intelligence is or may be, it all amounts to mental bukkake for all of its practical use.

We can dig ourselves even deeper down the rabbit hole if we’d like. We can assume a computer can not only be smarter than us, but also more capable, as a foregone conclusion. But once we do, we start to realize that our speculative reality has an innumerable amount of dependencies. Who is to say that tactile robotics will progress at the same pace? What if politics end up ruining the progress? So many things would have to go according to plan. The complexity needed to replicate human intelligence, while somewhat plausible (only because we have convinced ourselves on arbitrary evidence that it can be achieved), requires us to develop in 20 or 50 years what 3 billion years of evolutionary trial and error has granted us through a highly random and improbable path.

Our intelligence is more than just our brain structure; our bodies themselves and our individual origins shape how we interact with the world. While in no way do I believe in some unreproducible spirit that drives us, I feel that even our ideas of what a perfect artificial intelligence are will run into far more barriers as we progress than we can now anticipate. But then again, that is speculation on my part, just as the whole notion is to begin with.