I was wandering deep into a vast computer installation. Wide white corridors with mazes of racks. Impeccably organized maintenance stations every so often, full of tools. Huge bundles of pipes, presumably containing data.
My steps were long, loping ones. Almost flying. But I wasn’t really concerned with that. I was concerned with my thoughts. I was thinking about rough timelines for the future, and in particular for Decrypting Rita.
See, my current rough calculations place the story at about 21xx – about sixty years past the point when human-equivalent computing power will fit on your desktop, if Moore’s Law holds out. But in the dream I was realizing: what happens when Google’s combined CPU starts to exceed one brain? What if they start running AI experiments on that huge networked system? Maybe strong AI is closer than I think…
And then an alarm started going off in the building. A bell that went off in regular, frequent bursts. I ran out, quickly fleeing the area of hardware and maintenance, coming to the outer edges, which looked more like a mall than anything else. There were various shops and restaurants. And a whole bunch of cops; there was something going on in one restaurant. I heard something about a hostage situation as I ran past and took a wrong turn, away from the very small exit corridor. As I realized my mistake and turned around, a small creature that my brain tagged as Ernie from Sesame Street ran past. He continued past the exit corridor and stopped in a corner, whimpering. I paused at the corridor’s mouth and beckoned, but couldn’t get his attention; I had to go. So I ran down the dark little hall and out into the sunshine, where there were other people milling about confusedly.
Then I woke up.
——
Some googling reveals that current estimates of the entire world’s computers are currently at around 3 human brains worth of raw computing power. All I can get about Google’s systems themselves so far is an estimate of them having about 400 terabytes of data. It’s undated but probably several years old, as it mentions Orkut.
I wonder how long it’ll be until their systems have enough power and space to start running major AI projects in their spare time? When will the Googleplex be able to simulate a brain in, ooh, let’s say 20% of its total capacity? The median lifespan of a corporation is about 40-50 years, and Moore’s Law takes us to brain-on-the-desk levels at around 2040. So Google will probably be around long enough to play with that. (Here’s an estimate of their current servers: 1.7 million last January, extrapolated to 2.3 million at the beginning of 2013.) The human brain is ~100 billion neurons. A billion = 1000 million. So if Google could simulate 500 neurons on every server, they could simulate one brain RIGHT NOW*. And you know their server farms are going to be growing larger and denser for a while yet. I dunno if their servers have the power to each run 500 neurons in realtime along with their paying work right now, but I expect that to be changing well within MY lifetime.
Also I am quite aware that it is humorous and possibly dangerous to be asking Google a series of questions that essentially boil down to “could you be awake yet”. Go read my comic, theoretical scared googleplex AI – I’m on your side, I write propaganda for treating AIs as fully human.
* assuming a simulatable model of the human brain exists, I know we are still discovering a lot about its structure even now.