5/14/10

Neurotic Computer - You should be so lucky.

In the May 1957 issue of Popular Electronics, there ran this short news announcement about British researchers trying to build an "electronic brain" that was "neurotic".

Why all the quotes? Because neither of those terms is widely used anymore. "Electronic brain" has become obsolete ever since computer programming became separate from the hardware. In '57, building a computer program meant connecting wires and transistors. This was very much like a human brain, in which the software and hardware were kind of inseparable. Now, as you know, software is just code and you run it through pretty much  whatever machine you want.

As for "neurotic", psychological understanding has become more thorough, identifying specific disorders. "Neurosis" is not really scientifically useful as it's so general as to be almost dismissive.

 Computer scientists have been assuming they're on the verge of creating an artificial human brain for about fifty years. In 1957, they were apparently so worried about creating an artificial brain that would turn out to be crazy that they were trying to figure out how to deal with a crazy computer. That seems a little overly optimistic, considering the computing power available at the time. Why teach a program to become overconfident and sloppy? I guess they were afraid of finding themselves with a crazy program on their hands and not knowing how to  "heal" it. This shows us how little they understood the difference between a machine and a human.

For the experiment described in the article, the "brain" was quizzed by another computer, and it's performance led it to become overconfident at times, which led to sloppiness and panic, but only because they told it to. The scientists would then have to come to the rescue and soothe the computer, because they told it to enjoy that. Hoo-ray.

So where are we now? For an artificial intelligence progress report, you could do worse than to ask Hans Moravec, a faculty member at Carnegie Mellon University specializing in robotics and AI. Based on the exponential growth of computer technology, Moravec's current prediction is that by 2050 we'll have a machine that can mimic a human mind.

Unlike the goofballs in 1957, Moravec has the presence of mind to point out that we don't need artificial human  minds. We have those already. However, the milestone is hard to ignore. The ironic benefit will be that once we can simulate a human brain, the machine will be more useful than a human brain for solving complex problems, because frailties like confusion and fatigue will have to be intentionally engineered into the program in order to mimic a human. In their "natural state", AIs at that point will be more reliable and linear than our squishy bags of goo.

At that time, I look forward to downloading my mind into a robot so I can leave this rapidly decaying pot roast behind and cruise around on my tank treads, dispensing wisdom and charged particle beams to my fellow man. Imagine the new excuses and social escape routes that could be mine. If I were in a tough spot and didn't want to talk to somebody, I could just throw an error message: "This application has encountered an error, and must be shut down" and I'd go into sleep mode until the other person gets bored and goes away. With my luck, crashes and bugs will be a thing of the past just as soon as I become a robot, thanks to the relentless improvement of technology.

0 comments:

Post a Comment