Idea of the Day #22
Super Super Human Intelligence
Thoughts on artificial intelligence that outsmarts it’s own creators
Whenever I’ve imagined artificial intelligence, I have been making 2 crucial mistakes. It was reading the book Superintelligence that helped me realize these flaws in my thinking.
Realization #1: AI May Become Smarter than Us
The first realization came straight from the book: I’ve always assumed the AI will be as smart as humans. I had never considered the possibility that AI would surpass the intelligence of humans.
But let us note at the outset that however many stops there are between here and human-level machine intelligence, the latter is not the final destination. The next stop, just a short distance farther along the tracks, is superhuman-level machine intelligence. — Bostrom, Nick. Superintelligence (p. 5). OUP Oxford. Kindle Edition.
I am not sure why I never considered this “end” before. The fact that AI will not “end” with human level intelligence (if it indeed makes it there). I suspect it has something to do with fear. I have never undertaken any kind of investigation of AI, even though I once spent a semester “doing it” with material science data.
I can’t put my finger on what about AI makes me uncomfortable, though there are all sorts of perfectly logical reasons to be put off by such an idea. Perhaps it has something to do with the nature of knowing and knowledge. I’ve been considering the subject for a while now and it, too, makes me uncomfortable in a strikingly similar way.
We are not rational creatures, which is bothersome enough, but what if we coexisted with more intelligent creatures who’s inner workings are opaque? Perhaps that is where my discomfort stems from. While my time with AI on that project a while ago was limited, it was enough to help me understand that it’s very hard to understand exactly what an AI does to reach conclusions.
From my perspective, before reading this book Superintelligence, I perceive AI as lacking transparency. Who knows what goes on inside of those artificial neural networks?
Realization #2: AI May Be Smarter Than Us in a Different Way
Following this line of discomfort is the idea that AI could become smarter than us in ways totally foreign to us. It is certainly possible that there are others ways of reaching knowledge and conclusions than we are aware of.
For me personally, it would be quite frightening to be around a machine that is not only smarter than me, but completely foreign.
In the first lines of Superintelligence, the author presents an interesting question: How do we control AI?
It’s my second realization that causes me concern when asking this question. How can we control something we don’t even understand?