A Cute Gimmick

A couple of people have suggested that I should stop bombarding y’all with rants against the DMCA and focus on something else for a change. To that end, I’ve begun reading Ray Kurzweil’s The Singularity is Near, a book about the coming technological super-race that got a lot of buzz last fall, and has been sitting on my bookshelf waiting to be read since then. Over the next few weeks, I will instead bombard you with rants (and possibly raves) about that.

I’m skeptical of his basic claim (which I won’t re-hash here–check out the Amazon link above if you want a good summary) both because I’m suspicious of utopian claims in general, and because strong AI is a problem that has consistently bedeviled computer scientists. Since the 1960s, computer scientists have been predicting that human-level artificial intelligence is just two or three decades away, and so far the goal seems to be getting further away as we learn more about the difficulties involved. Which isn’t to say we’ll never get there, but I don’t think you can simply extrapolate Moore’s law out to the point where computer chips have as many transistors as brains have neural connections, and assume that at that point we’ll have human-scale intelligence.

But I’d like to start with a more specific nitpick, which I’ll give below the fold…

This chart is the centerpiece of chapter one:

The basic idea here is that in each “epoch,” the time between major paradigm shifts gets exponentially shorter. So it took (say) a billion years to get from the start of life 2 billion years ago to the start of eukaryotic cells a billion years ago. But it only took about 10 million years to go from humans walking upright to the evolution of stone tools about 30 millions years ago. And it took about about 50 years from the invention of electricity 150 years ago to the invention of the computer about 100 years ago. (I’m just reading these numbers off the chart, so the precise values probably aren’t right) Each paradigm shift, he claims, is happening faster than the one before it.

Now on some level, this is obviously right. Human technology works faster than evolution, and recent technological progress has been faster than it was thousands of years ago. But the chart proves less than he seems to think it does. In the first place, a log-log chart is extraordinarily forgiving. If you double or halve any variable in the chart, it only results in a slight shift in the visual position of that data point. Thus it’s relatively easy to get something that looks like a straight line from a log-log chart.

Then there’s the problem that Kevin Drum spotted last September. He extrapolates the chart out and calculates that if Kurzweil is right, then last month we should have seen a major innovation about once a week. Web 2.0 hype aside, that obviously isn’t happening. Kurzweil responds, with some justice, that it’s not fair to extend the chart out as he’s just trying to show the general past trends, not find a rigorous mathematical model suitable for extrapolating into the future.

But while I think his reductio ad absurdum is a little unfair, I think Drum has spotted the fundamental problem with Kurzweil’s chart. Kurzweil, after all, titles the chart “Countdown to Singularity.” He clearly intends to give the impression that there’s something special about the present (or at least the next 50 years or so). But if you think about it, the seeming linearity of the chart has more to do with the limits of human cognition than with anything underlying trend.

To see why, imagine that it’s 1906, instead of 2006. I’m creating the same chart. I would probably list:

Printing Press: 450 years ago
Steam Engine: 200 years ago
Railroad: 100 years ago
Telegraph: 70 years ago
Telephone: 30 years ago
Radio: 12 years ago

Looks to me like humans were going to transcend biology around 1945! Perhaps the A-bomb was the 20th century’s singularity.

I could do the same thing for 1956, for 1856, or for 1806. A medeival historian could probably do something similar for the year 1006. When we look back in history, we tend to know and care more about things that happened recently than we do about things that happened a long time ago. Off the top of my head, I can name about a dozen things that happened in the world between the year 0 and the year 1000. That’s not because only a dozen things happened, but simply because I don’t know nearly as much about it, on a per-year basis, than I do about things that happened more recently.

The same principle applies over longer time horizons. A human being being living in the lower paleolithic era probably would have felt that a lot of important things had happened in recent centuries. He might even have been right. But because we’re so far away from it, we neither know nor care about most of them. We know that human beings invented some stone tools, made some cave paintings, invented fire, etc. But we probably don’t know about a lot of smaller inventions that–at the time–would have seemed like a big deal, but that we now either have forgotten or take for granted.

Hence, the seeming linearity of Kurzweil’s chart is largely a function of the fact that we tend to learn less about a given time period the further back it is. That doesn’t necessarily prove that anything profound is going on. Of course it also doesn’t prove that it’s not–I’m looking forward to reading on to see what other evidence he has.

Update: Now that I think about it, why does he get to lump “telephone, electricity, radio” into one event? Each of those alone is arguably as revolutionary as the computer. Yet if we made those three separate events–spaced out by about 15 years each–it would ruin the nice linearity of his chart. In fact, you can argue that the pace of invention has been pretty much linear since the industrial revolution, with a major invention every 15-20 years. The second half of the 19th century saw an explosion of new inventions that arguably hasn’t been rivaled since, at least in terms of their transformative effect on society.

SHARE: