Within thirty years, we will have the technological means to create superhuman intelligence. Shortly after, the human era will be ended.—Vernor Vinge

You are to be forgiven if you have yet to hear of the coming singularity. It’s a science fiction premise espoused by SF author Vernor Vinge back in 1993. Boy genius Ray Kurzweil put meat on the bones of the idea by writing multiple tomes on the topic. The premise of The Singularity is that soon, perhaps within one to four decades, we will be able to build machines with something rivaling human intelligence. Shortly after that happens, the age of humans will end as machines evolve like...well machines, and leave us to choke in their dust. Of course, that’s not how Vinge and Kurzweil see it. They’re optimistic that the machines will serve us. Or at least tolerate us. Apparently, they haven’t seen the Terminator or Matrix movies.

In any case, I attended a lunch-time interview with Intel’s CTO Justin Rattner at the Computer History Museum today. The interview is part of a year-long series of events at the museum, which is celebrating the 50^th year of the Integrated Circuit. (That’s a big thing here in Silicon Valley.) During the interview, Kate Greene, Information Technology Editor at the MIT Technology Review gently tossed softball questions at Rattner. Most of the questions at the beginning focused on the singularity.

Greene’s first question concerned when we’d know that the singularity had arrived. Rattner replied that we’d know it was here when we saw a robot emptying our dishwasher. In other words, when we’ve handed routine tasks over to machines, then we should know.

Then there’s the problem of getting the machine to be “comfortable” while working around humans. At first, I thought Rattner meant the embarrassment a mental giant of a machine might feel while performing such menial labor (think Hitchhiker’s Marvin the melancholy robot) but he actually meant being able to build a machine that was aware enough of its surroundings so that it wouldn’t accidentally stick an errant (but clean!) fork or steak knife into an innocent, fleshy bystander. Sound’s like a very good plan to me. I don’t plan on being an early adopter of the dishwashing robot.

Now I’m writing this blog entry with tongue firmly planted in cheek, but one thing I certainly agree with Rattner about: technological progress is accelerating at an ever-quickening rate. Rattner alluded to the bird-bone flute discovery—just announced today—that was found in the Ach Valley of southern Germany. That means that scientists now have a record of human artifact development that goes back at least 35,000 years or about 30,000 years before the flood. Rattner says that we will see more technological development in the next 100 years than in the previous 35,000 that is, if we (or the robots) don’t kill off the human race in the next 100 years. Even if we or the robots do kill us off, the prediction probably holds because the robots will do it anyway. (Note: Some people actually think that Asimov’s three laws of robotics are real and don’t fear robots. My father, a personal-injury attorney, had cases where very dumb robots killed people. Sorry Dr. Asimov, positronic brains seem in short supply at the moment.)

Turning to some actual technology, Rattner predicted that the emergence of machine intelligence would lean heavily on the development of statistical and probabilistic programming techniques. That way, even if the software isn’t 100%, the machine can still function. We cannot expect perfection from machines and we need to shoot for 95-98% accuracy in the high 90 percentile. We need to be able to live with that lack of perfection too. After all, that’s pretty much all we can expect of ourselves and that’s working at our best. When things like nuclear weapon launches are involved, we know we need to build in some redundancy or we’ll wipe ourselves out. In a less intense example, we all rely on automated spelling checkers these days to help us with imperfect spelling. 

Rattner also displayed a very practical attitude towards semiconductors. We got lucky at the most recent crisis point. Silicon-gate CMOS died quite quickly but at exactly the right time, fabrication geniuses had High-K, metal-gate CMOS at the ready and process technology advanced without a hiccup. Next time, we’re not getting off so easily. We’ve got three to five generations of process technology left with CMOS as we know it and then charge-based, solid-state silicon electronics is finished. We’ll need to switch to spintronics, carbon nanotubes, or graphene to maintain Moore’s Law.

Finally, back to the singularity. Today, as I’ve said before, we build toys. “My iPhone knows nothing more about me today that it did when I bought it,” quipped Rattner. It’s true. Today’s machines are woefully ignorant about their environment. That’s because sensors are still expensive. IC vendors have made inroads and have developed some semiconductor-based sensor technologies, but the whole field remains in a primitive state. More grist for the technology mill. See you at the singularity.