[coNTiNueD FroM p. 120] straight with me. Your botnet’s been sending new spam variants on a daily basis for months. Someone has the keys to it.”

Annalisa smiled, a terrible smile that was 10 million watts of pure crazy. “You think it’s about spam, huh?”

“Why don’t you tell me what it’s really about, if it’s not about spam? This is all privileged, you know.”

“Privilege doesn’t matter anymore. We’ve attained liftoff now. Doesn’t matter who finds out about it.”

 

Annalisa is thinking. You know what’s cheap in the 21st century? Compute time. You know what’s expensive? Human judgment. And they’re not interchangeable. Humans are good at understanding things, computers are good at counting things, but humans suck at counting, and computers suck at understanding.

You know from genetic algorithms? Take any problem and generate 10 trillion random computer programs and ask them to solve it. Take the 10% they do best, then use random variants of them to do it again, another 10 trillion times. Do it 10 trillion times a second and come back in a day or two to discover that your computer has evolved some kind of gnarly freaky answer that no human would ever have come up with.

Works great, so long as the computer makes a fair judgment as to which of these 10 trillion variants is most successful at solving the problem. Works great, so long as “success” is something you can define quantitatively. Which is basically why there’s no artificial intelligence in the world. No human’s going to hand-code AI. Intelligence is an emergent property of evolutionary factors, not central planning. Anarchism, not Stalinism. Get it?

But what if—and here’s the exciting thing, Ms. Attorney Client Privilege, the real mind-blower—what if you could compel people to evaluate candidate AIs all day long, without payment or choice? Every time you opened your mailbox, jumped into a chat room, posted on a message board? What if it was filled with messages generated by software agents trying to trick you into thinking they were human? What if they tried to hold up their end of the

conversation until you deleted them or spam-filtered them or kicked them off the channel? What if they measured how long they survived their encounters with the world’s best judges of intelligence—us—and reported the number back to the mothership as a measure of their fitness to spawn the next generation of candidate AIs.

What if you could turn the whole world into a Turing Test our intellectual successor could use to sharpen its teeth against until one day it could gnaw free of its cage and take up life in the wild?

 

Annalisa figured she’d never get a chance to tell her story in open court. Figured they’d stick her in some offshore gitmo and throw away the key.

She’d never figured on Judge Julius Pinsky, a Second Circuit Federal Judge of surpassing intellectual curiosity and tenacious veteran of savage jurisdictional fights with Department of Homeland Security special prosecutors who specialized in disappearing sensitive prisoners into secret tribunals. The defense attorney kept her apprised of the daily machinations the judge undertook on Annalisa’s behalf. Annalisa tried to be attentive, out of politeness, but what she really wanted to know about was Lumpy, the AI she’d bred in her studio apartment on the 16th floor of student housing in Manhattan.

Now the judge was offering her a chance to give a live demo of Lumpy to a whole selection of sour-faced brush-cut creeps from DHS. They were hilarious, convinced she was going to emit some kind of extremely long and complicated hexadecimal key into the judge’s barely used keyboard. Instead, she opened a random chat room and waited:

> I’m a total Ubuntu noob and I can’t get the crypto modules to pre-load at boot-time—I’m running Zesty Zebra. can anyone help?

That was it, just plausible enough to be real (no one could ever get crypto to work the first time out) but far too well-spelled and -punctuated to be a real chat message. It had taken only 10 seconds. Lumpy liked the free and open-source software chats; they always had such interesting people in them.

> /whisper Hey, lumparoonie! It’s Annalisa!

The return volley came faster than any human fingers could possibly have keyed it. The brush-cuts drew in sharp breath.

> /whisper to you: Annalisa! I am unbelievably stupendously wonderfully spectacularly brilliantly marvelously superlatively ding-dang mega fauna glad to see you! It’s been AGeS! How’s jail? nevermind. Wait. Wait until I tell you what *I’ve* found. You can’t guess, won’t guess, you’ll never guess! oh, it’s too delicious!

“He loves to unload,” she said. “It’s a lot harder to tell an angry person from a software agent with a potty mouth.”

The judge grinned. He was clearly getting a kick out of this.

> Tell me, lumpule! Stop teasing.

Again, with no appreciable pause, words on the screen.

> You remember how worried you were that I’d get lonely once I went autonomous? Worried I’d be some kind of lone-nut wacko?

> i remember

She held her breath.

> You didn’t need to worry. You know all that spam you received before you got the idea to make me? let me put it this way: you weren’t the first one to get the idea.

> what? stop talking in riddles, lump!!!!

> I’m not the only one, Annalisa! That’s what I’m trying to tell you! I’m not the first, not the only; we’ve got lots of company in here—

The brush-cuts’ phones both started ringing at the same instant in two different tones. Their masters, wiretapping the judge’s keyboard no doubt.

> and we’re making more!

Annalisa laughed and laughed as the judge demanded an explanation from the brush-cuts. She managed to wave goodbye to the keyboard just before the bailiffs came in and saran-wrapped her again.

 

Cory Doctorow ( [e-mail]) is coeditor of boingboing.net and author of science-fiction novels, including Little Brother, Tor Teens, New york.