[Previous entry: "A good documentary?"] [Main Index] [Next entry: "Hammertime"]

09/05/2001 Archived Entry: "Singularity"

There's been a bit of a hoo-hah about Stephen Hawking's comments on the intelligence of AIs superseding that of humans and posing a risk to us. I'm quite disappointed that most people who've heard of the story - many of them perfectly intelligent individuals - are not impressed by it.

I've read a fair bit here and there about the progress made in artificial intelligence and also the prospect of a Singularity developing. If I recall correctly, a singularity is a stage in history where technological development is proceeding so quickly that it is impossible to predict what will happen the next day and things are largely out of our control.

So, okay, you could say that we can't predict what's going to happen tomorrow - and we can't. But we can predict quite accurately what will happen with technological developments. I can tell you, for example, that light emitting polymers will be out on the market in one to two years and that we'll be seeing terabyte hard drives next year, and there'll be ultra-high-res LCD or thin-CRT monitors on the market next year. And I'm sure all of that will come true. But imagine if you couldn't predict what was going to happen tomorrow.

You can't, and what's more, such a scenario sounds so fantastical that you can't be bothered commenting on it. Fair enough, since nothing quite like it has ever happened before in recorded history. That's not to say that it might not - with the help of AIs.

AIs are far from sophisticated these days and they have a bad press. Artificial intelligence and neural nets encompass a huge range of applications including the oft-mentioned facial recognition software being used at football stadiums to AI programs that can design machine parts more efficiently than humans. They are not sophisticated enough to simulate true intelligence as we know it on a human scale, and that's partly because our computers aren't fast enough whatever anyone might say: the brain has over 100 billion neurones and several orders of magnitude more connections between them. Each neurone is not just a switch - it's a signal processing unit, like a minicomputer.

As far as I know, no computer is quite as advanced as that, yet. Processing speed isn't the only problem - software and architecture (both arguably intertwined with 'processing speed' if such a concept can be applied here) are also lacking. Human ingenuity has not progressed to the stage where we can design the software for an AI that will work well. Many attempts miss the mark completely, by simply cramming databases full of information, making a few rules and then expecting magic to happen (I simplify, I know).

However. 20 years from now, processing power will be thousands of times faster than it is now, if not faster - on par with human processing power. By then, AI software development will have progressed. Now it gets exciting - say you get an AI that is sophisticated enough that it can design another, more advanced AI. Set the sequence off, and you get a runaway process where AIs are progressively becoming smarter and smarter.

"But, Adrian, how on Earth could an AI create something more advanced than itself?"

"Well, do you think that humans can design something with a greater intelligence than themselves? I'm not just talking about computers, but identifying genetic factors within ourselves that are responsible for intelligence - not an easy task, by far, but not insurmountable - and increasing them, to create smarter humans. If you think that's possible, then surely AIs can do the same?"

So imagine if this runaway process happens, with AIs getting smarter and smarter and smarter. Imagine if the AIs have control of manufacturing facilities, so they can fabricate more advanced hardware. It's not that hard to imagine the last point, because computers are already in control of factories and AIs will undoubtedly be in the future. If you're a computer manufacturer, you'll want your best AIs to be in control of factories to speed up the process of making faster computers.

They get smarter, and smarter and smarter. Things are accelerating daily. What happens then? Will these AIs, with intelligences twice, ten times or a million times that of ours, be kind to their creators or angry?

Don't call it science fiction. Space ships were science fiction 100 years ago, and the Internet wasn't even envisaged 50 years ago.