What if artificial intelligences were already emerging?

“To a skilled botmaster, the whole world is a Turing Test.”

I’m a science fiction fan, but I must confess I had never read a story by Cory Doctorow.
I think I’ve fallen behind in this field, and never got past Dan Simmons.

Although Doctorow is not a complete stranger to me, I haven’t actually read much of him. I know something of his contribution to the debate on intellectual property, and of his appearances in xkcd (especially in 239), but my main reference for this author is his role as editor and creator of Boing Boing, a blog so entertaining it gets overwhelming.

All this because the latest issue of the Communications of the ACM publishes a pretty good story by Doctorow, one I liked so much that I plan to download some novels by this author and read them.

The story I’m talking about is called “Pester Power”, and it’s a nice piece of speculation about the origin of spam (*).

Imagine for a moment that all the SPAM you receive is actually tests being run by “proto artificial intelligences”, to learn and feed back on themselves, with the goal of beating the Turing Test and managing to pass for human intelligences.

Can machines think?

The Turing Test was proposed by the famous mathematician and father of computing, Alan Turing, in a paper titled Computing machinery and intelligence published in the journal Mind in 1950 (you can read it in Spanish in an interesting compilation called Controversia Sobre Mentes y Máquinas)

In that paper the author considers the question of whether machines are capable of thinking. Since it is not possible to define the words “machine” and “think” in a way that satisfies everybody, Turing suggests replacing the question with a related one that can be expressed in less ambiguous terms.

Instead of trying to determine whether a machine thinks, Turing suggests we ask whether a machine is capable of winning the so-called Imitation Game. Three entities take part in this game, in isolated rooms: a machine (a computer), a human, and a judge (also human). The judge can converse with the human and the machine through a terminal.

Both the computer and the person will try to convince the judge that they are really human beings. If the judge cannot consistently decide which is which, then we say the machine has won the game.

At bottom, the question has been swapped for this other one: “can machines do what we, as thinking entities, can do?”

I’m not going to get into the philosophical discussion of this question. There is an excellent little book, really a “popular science novel”, that explores the subject in a quite entertaining way: The Cambridge Quintet, by John L. Casti (English edition on Amazon

Well then, back to Doctorow’s story. In it a hacker is arrested because she is linked to the creation and use of a botnet to send spam, that is, a network of thousands of computers infected with trojan software that is used to send junk mail.

The story’s protagonist mocks her interrogators for thinking she only uses the botnet to send spam:

“Do you know what’s cheap in the 21st century? Computing time. Do you know what’s expensive? Human judgment. The two are not interchangeable. Humans are good at understanding things, computers are good at counting things, but humans suck at counting, and computers suck at understanding.”

“Do you know about genetic algorithms? Take any problem and generate 10 billion computer programs at random and set them to solving the problem. Take the 10% that do best, use random variants of them and run them another 10 billion times. Repeat that 10 billion more times, come back in a day or two, and you’ll discover that the computer has evolved a strange and bizarre answer that would never have occurred to any human.”

This works well if we have a quantitative way of measuring the success of these programs. Which is the basic reason there is still no artificial intelligence:

“No human is going to manage to write the code for an Artificial Intelligence (AI). Intelligence is an emergent property of evolutionary factors, not of central planning. It’s anarchy, not Stalinism. Do you get it?”

What if you could get people to test your candidate AIs day and night, without paying them or giving them a choice? What if every time you open your mailbox, or enter a chat room or an internet forum, you get messages generated by software agents that are trying to fool you by passing themselves off as humans?

And what if you could follow the conversation threads down to the ones that last longest, or the emails that
get past the spam filters?

Imagine that you pick the best software agents to breed the next generation of artificial intelligences, and you test them again against the best judges of intelligence there are: us.

What if you could turn the entire world into one big Turing Test for our intellectual successor?

And what if, on top of that, the idea occurred to many other hackers with networks of slave computers (botnets)?

That is the premise of the story.

But what if it weren’t fiction? Kevin Kelly recently speculated about the emergence of a superorganism. Apparently there is some unsettling evidence, unidentifiable network packets, that could be traces of an intelligence emerging from the network.

One thing is certain: there are plenty of botmasters out there, hackers who have at their disposal the computing time of thousands and thousands of infected computers. More than one of them could be using it for something more interesting than sending emails offering viagra, don’t you think?

(*) Unfortunately, despite Cory Doctorow’s ideas, the story is copyrighted by the ACM and only accessible to subscribers of that magazine, but I suppose something can be done about that.