Sunday, July 22, 2012

From Stupid Computers to Smart AIs

We've all had computer problems, the misbehaving, lagging when there's no reason to lag, crashing at inopportune moments... What we have trouble remembering is that computers are very stupid devices: they only do what we tell them to do, and they are only as smart as the least intelligent programmer who built them.

Artificial Intelligence, using a very sophisticated computer as a brain, has long been a concept for science fiction writers--who are, on the whole, mildly prophetic. As computers and data storage in this unimagined world we live in improves, true artificial intelligence is becoming more and more possible, but it's still not something we're completely capable of. Yet.

The first barrier is storage. I'm not going to quote a number, because it probably changes depending on the source, and some of my viewers I'm sure would love to nitpick. I'm just not going to give them the satisfaction.

The second barrier has to do with one of the problems I mentioned above: computers are only as smart as the stupidest (or perhaps more appropriately, the least smartest) programmer. To explain a work-around for this, I offer another concept that has long been on the minds of science-fiction writers and readers, but far closer to conception: self-replicating machines.

For those of you who are not sci-fi aficionados, let me explain the concept. Consider a manufacturing plant that takes raw material into machines. Now those machines that it builds are identical in every way to the ones it uses. It also builds the walls and the ceilings, processes the concrete that makes up the floor, etc. In short, a plant that not only manufactures itself, but builds itself. I say this is closer to reality for two reasons (and I'm trying very hard not to get too far off topic): first, because it requires no actual intelligence to build, only a very detailed set of rules and instructions, and secondly, because it already exists on a smaller scale. I was browsing through 3-d printers a few months ago, and stumbled across one that was capable of building its own components. If it had any limitations, the computer chips were it.

Bringing this back to the concept of AI, the work-round for a computer being limited in ability by the skills of its programmers is to give the computer the ability to write itself. This removes any poorly-written, inefficient, or unnecessary code, and allows the computer to write better routines for the functions that it needs. This is the ability to learn. When we learn something, our brains change the way we look at things, or perform certain tasks, and an artificial, digital brain needs the same skill.

The third barrier is a willing builder. Too many people have been frightened away from the concept by popular media, from "The Matrix" to "I, Robot" (the movie, not the book). Many more examples are out there, but I feel no need to continue the list.

Defeat those three barriers, and creating true Artificial Intelligence is practically a walk in the park. Just make sure it's an actual AI you're trying to create, and not just a computer designed to pass the Turing Test.