Duncan's Reflections

Hi everyone! I doubt anyone outside of my class reads this (in fact, with what someone in class said today in mind, I doubt anyone outside of Professor Waldo and Dean Smith read this), but if you exist, I didn’t post last week because Columbus Day is a Harvard holiday, so we had no class. But now we’re back and better than ever, moving into artificial intelligence the Singularity.

If you’re unfamiliar with the Singularity, think about it like this: humans manage to create superhumanly intelligent beings. These beings create beings who are even more intelligent and who create beings… I hope you see where this is going. The resulting intelligence explosion is commonly called the Singularity, as it will propel us into a completely new era, one where human intelligence is virtually obsolete.

In class, we had a big debate about whether we could create a superhuman intelligence that was like a human in every aspect. I argue it’s possible (there’s nothing magical about the brain) but it seems like an absolutely awful idea. One person in class raised the point that we worry that more intelligent beings will want to subjugate us, just as we have subjugated all other life on this planet. He continued to ask, “Does that speak to the beings we will create or to ourselves?” Maybe that’s what we fear because that’s what we do. And in that case, why in the world would we want to create a human-like super-intelligence? It would be forging our own shackles. Perhaps more intelligent beings will necessarily subjugate less intelligent beings. But we KNOW human-like intelligent beings do that. Why would we create them? People often try to explore new technological territory by doing something that is familiar with it, but with the stakes this high, we can’t afford to mess up.

Of course, if we’re not creating human-like intelligence, what exactly is “intelligence”? I would define intelligence as the ability for an entity to function on its own, conducting various input-to-output functions that self-optimize as the entity takes in more input. For example, as humans we take in input (stove is hot) and produce output (take hand off stove). In the future, our decision-making functions change to prevent us from putting our hand on the stove in the first place. You could take issue with this definition—indeed, approximately half of class today was spent trying to nail down a definition of intelligence. But I like it. Intelligence in this way is not restricted to things like playing chess or solving math problems. It can apply to social interaction, emotions, literally anything you can imagine that an intelligent being would have to do. It also seems to fit well with ideas of “machine learning”, about computers changing their algorithms based on the input they get.

Of course, this definition opens the door to discussion of free will because if you believe in free will, you probably don’t accept my idea of decision-making, empathy, emotions, etc. as just input-output functions. But they certainly seem to be. The human brain is just a collection of neurons firing, or on a lower level, just a collection of chemical reactions. There is nothing magical about it. Which isn’t to say it isn’t a beautiful piece of machinery, just that it’s deterministic. If you put in some input, you will get a pre-determined output (note again that I don’t think recreating human brains is good—this is just to give an example of a well-known intelligence). So while free will is a white lie I tell myself to make my head hurt less when grappling with decision theory, it is certainly not true and shouldn’t enter discussions of intelligence.

So now we have this definition of intelligence. How do we build God with it? How do we not mess this up? I don’t know, but I believe these will be the most important questions of our generation.