On Self-Improving Intelligence

by Malte Skarupke

I just watched this TED talk by Nick Bostrom which is about what happens when computers get smarter than we are. It follows the theory that once we have created AI that is smarter than us, that AI will be able to create AI that is even smarter than itself. So at that point we’ll be left behind very quickly. Intelligence will develop a dynamic of self-improvement that can’t be stopped.

What he doesn’t realize is that we are already in that mode. We have already created people that are smarter than we used to be. And those people have created people who are smarter than themselves. That process has been going on for hundreds, if not thousands of years. The human brain is not the limit for human intelligence.

In his talk he has an image of what most people envision the scale of intelligence to be:


Where at the one end of the scale you have a stupid person, and at the far end of the scale you have a really smart person. In reality, Bostrom argues, it looks more like this:


Where there isn’t much difference between a chimp and a human, and there isn’t much difference between a stupid person and a smart person. After all if you look at the biology of a chimp and the biology of a human, not much evolution has really happened. And biologically speaking the difference between an idiot and a smart person is tiny. Thus the gap in intelligence is tiny. This interpretation is not true.

An unaugmented human wouldn’t be much smarter than a chimp, true. I would draw the circles closer than they are in his picture. However with technology humans are vastly smarter than chimps. As a matter of fact the difference between humans is enormous, depending on what technologies they have mastered.

What technology makes people smarter? It’s things like writing. Ever since writing was invented, we could transfer knowledge more reliably from one generation to the next. That easily added 10 IQ points to human intelligence. Roman numbers also added something like 10 IQ points to humans: With roman numbers you could add numbers that were far too complicated to keep in your head. This allowed forms of organization that were previously impossible. Arabic numbers added another 10 IQ points. In fact they made us so much smarter that almost everyone could learn to multiply. (where before you had to learn a weird base 60 system)

There is a great paper that I can’t find any more which talks about how Galileo derived a law of nature and drew some very simple conclusions from it and how complicated this was for him because he didn’t have equations. Once equations were invented, every high school kid could do what Galileo did. I wish I could find that paper to link to it…

That doesn’t even cover a small part of the technology that we invented to make us smarter, and so far I have only talked about the middle ages and before that. So let’s jump over all of history and talk about modern times. Computers and the Internet have given us huge boosts again. (and I’m obviously over-simplifying when I say “Computer” or “Internet.” The Internet kinda sucked before the World Wide Web. The World Wide Web kinda sucked before Google. Google kinda sucked before Web 2.0 made it so that people actually created content like the Wikipedia. Many steps of improvements happened here and many more will happen. And people who used the Internet before the World Wide Web already thought it was the best thing ever) Smartphones make us smarter again. Just think of how much memory work you have outsourced to your smartphone. I remember almost nothing any more. It’s all in my smartphone. I just remember where I can look things up.

So yes, in theory humans aren’t much smarter than chimps. It’s tough to remember this, but if you had never learned a number system like the Arabic numbers that you probably use every day, you wouldn’t be able to add 24 and 44. It would either be impossible for you or it would take you many minutes and you’d probably get it wrong. We forget hat we use technology like ‘Arabic numbers’ to make ourselves smarter than chimps.

This process probably started when people first settled down, had governments, specialized and built cities. And then it got a second start with the printing press because that allowed us to make many perfect copies which is what allowed us to get out of infinite cycles of getting smarter followed by losing information to centuries of entropy.

We are nowhere near the end of that process. People now are much smarter than people were forty years ago. People in forty years will be much smarter than people are now. What are changes that will make this happen? I can only think ahead a small amount of time, but here are a few ideas. If programming was so easy that everybody could do it, that would make us a lot smarter. If smartphones and the Internet were easier to use, that would make us much smarter: If it didn’t take me thirty seconds to take my smartphone out of my pocket, unlock the screen, open the browser, type my search query into Google using a terrible on-screen keyboard and to then finally be able to click on what I’m looking for. If that whole process took one second or less, that would make us a lot smarter. If we had an easy interface to collect data, aggregate it, graph and map it, analyze it and look for outliers, so that everybody could do big data processing, that would make us a lot smarter.

And I’m probably not thinking of the big invention that will actually have a huge impact. But I genuinely believe that a better interface for programming or a better interface for data collection and analysis could make people smarter.

So we will obviously write real AI at some point. But we’ll also keep improving ourselves. And it could be that at some point improving ourselves will be much more difficult than improving AI. That doesn’t have to happen though. There is a possible future where computers move ever closer to our brains. From separate rooms to our desks to our pockets to our heads to our eyes and ears and finally into the brain itself. We’ll be running all of the AI routines that we are currently inventing. Just like I’m currently outsourcing my memory to my smartphone, maybe in the future I’ll outsource image recognition to an algorithm that runs in my brain all the time. I’ll have terminator vision that highlights the important things so that I can read a book while walking through the city. (or hey, my audio processing requires a lot of focus. How about I run a speech recognition algorithm so maybe I can listen to a few podcasts at the same time while also keeping up a conversation) And it doesn’t even have to be this complicated. The point is that we’re nowhere near the end of improving ourselves.

And even if AI does take off and leaves us behind, we will probably come up with a way to catch up. Because often all it takes is that somebody (something) shows you what is possible, and all of a sudden you will find a way to get there. Catching up is always easier than marching ahead, even for AI.