08 February 2010

Signs of the Singularity (II)

Yesterday, while eating appetizers before the superbowl, the question arose "what time is the superbowl kickoff?" I responded "Well, we could ask Google. Wouldn't it be funny if we typed 'what' and it suggested the rest of the question?". Since a terminal was nearby, I woke it up, started up firefox, and typed 'what' into the Google search box. Lo and behold, Google suggested "What time does superbowl 2010 start."

About a year ago (last April) IBM announced it was preparing a computer to play Jeopardy. Of course, Google has effectively been playing Jeopardy for a few years now. Jeopardy is basically a game of trivia recall. A few keywords in the form of an answer are designed to trigger an association in our brains which we have to phrase in the form of a question. The form of the trigger and the form of the response are minimally important. The key to the game is trivia recall. And Google is the pre-eminent trivia recall system.

So, for a mere billion dollars or so, we had a computing system that outperforms a human brain at a task that is far more about general intelligence than chess. This year, we have a trivia recall system that can guess your question before you ask it. This is very much unexpected, emergent behavior. The system is definitely showing signs of high intelligence.

But we've seen super-human intelligences for a rather long time. The Egyptian pyramids could not be built by a single person -- they were built by a group of people. Building European cathedrals involved large numbers of people over a century. Putting a man on the moon was a group effort of hundreds of thousands of people. So one more super-human intelligence is not all that interesting. Except that this one is completely non-biological.

Expect to see another interesting emergent behavior from the Google network in the coming year. I don't know what it will be, but if we're going to hit the singularity, the timing is required. Human language translation is the pre-eminent human task that is crying out for a robust solution. And we now have sufficiently large databases of actual human sentences in each language that computers can start to put together sentences that sound correct to native speakers. If we get that connected up to a user interface, that would be pretty cool. If we can start routing Skype data packets through a convenient Google data center close to the path the data packet would take, we might be able to translate spoken conversations on the fly.


Post a Comment

<< Home