singing to the infinite
kill squishy humans
I've recently attempted, with my puny and unenhanced human brain, to read Ray Kurzweil's The Singularity is Near. It's one of those books that walks straight up to the line between brilliance and madness, and hits it with a sandblaster until it's rubbed out.
Premise: the rate of technological change is accelerating. Kurzweil believes this, and he has a ton of graphs and charts to back him up. Because the rate of change is accelerating, and because most of us make predictions as though the rate were constant, our entire society has greatly underestimated how very different the future is going to be.
Kurzweil paints an alternate future: by 2045 (give or take a few years), artificial intelligence will no longer need us. It will be good enough. It will be smart enough. And doggone it, people will like it. Like other lines mentioned earlier, the line between artificial and natural intelligence will be blurred out of existence. Some of us will have our intellects augmented, some of us will be pure software, and the Amish will have their own cyborg armies.
Our bodies -- or at least the bodies of those of us who choose to keep them -- will be full of nanomachines which will provide the same services to our bodies, only better. They'll also repair our cells faster than living damages them, effectively rendering us immortal.
Most important, the brainpower available to society will have exploded exponentially. Hundreds of billions of artificial brains, working much faster than normal brains, and not having to waste their time with sleep and bathroom breaks, will drive ever faster acceleration. Technological barriers will crumble instantly, and the gap between what humankind can do and the limits of the physically possible will slam shut.
Then we will be gods, the living music which infuses the cosmos with understanding and meaning. It will be nifty.
But I'd just settle for the robotized blood cells that allow me to hold my breath for hours.
It's a compelling vision, and very coherently argued. He seems to have glossed over a few of the difficulties, but I think he nailed the two biggest ones. I think the bulk of his vision hinges on two factors: Moore's Law* has to hold for about forty more years, and we have to invent the ability to simulate the human mind in an effective way. The first one may be tricky. Using light to burn silicon paths on a chip can only get us so far. I've seen predictions that it will run its course by 2019, 26 years too soon. Unless something supersedes it, we'd be left with computers about 1/8000th as powerful as they would need to be.
The author, of course, thinks that a better technology is inevitable. He's probably right.
The second feat -- simulating a human brain -- may turn out to be the easier half of the equation. Our tools for observing and understanding the brain have been undergoing the same sort of exponential improvements that our computers have. By the 2030s, Kurzweil figures that the tiny machines in our brains will be able to record our brains in arbitrarily complex detail, down to the structure and firings of each synapse. This will allow us to copy brains and simulate them in software. The nanobots will also be able to cause and suppress brain activity, allowing our brains to experience anything, from virtual reality to the recording and playback of sensory and emotional experiences, to a form of telepathy with other brain-hacked people.
But most important is the ability to simulate brain activity in software. If I had to guess, I would say that Kurzweil is actually being conservative about how much processing power that would take. We could very well figure out how to simulate some of the brain's substructures in a greatly simplified way. But that's the thing about Moore's Law: regardless of whether the brain simulation is a thousand times simpler than Kurzweil estimates or a thousand times more complicated, it barely alters the timeline.
If we assume that Moore's Law continues to double the amount of computational power available every two years, a thousand-fold increase in the complexity of the simulation only pushes the result about twenty years (ten doublings) into the future. A thousand-fold simplification of the problem only pushes it twenty years closer.
The amazing thing is that Kurzweil can make such fanciful predictions without striking me as a kook. That is either a tribute to his insight and rhetorical skills, or evidence that I myself am also a kook, and that we kooks often sound quite sane to each other. His conclusions also fulfill some unmet emotional needs for people who don't believe in an afterlife. When faced with an idea that I would clearly want to believe, my second instinct is to temper my enthusiasm for it.
My first instinct, of course, is to believe it. I have no idea why Kurzweil hasn't used this as the basis for a new religion. He'd make a boatload.
I'm going to make my own, much more humble prediction: someday, your cell phone will double as a universal remote. It will be a golden age of humanity.
Update: Leon says the iPhone already does this.
* I mean the soft, fuzzy, bastardized version of Moore's Law, which says that computers keep getting exponentially faster. Look upon my misuse, ye pedants, and weep.