Wednesday, February 16, 2011

Artificial Intelligence is Gaining on Us (Jeopardy! part 2)

I watched the first 2 episodes of the Jeopary! challenge between former champions and an IBM super-computer. Humanity isn't fairing well.

We may live long enough to watch the debate in Congress over whether computers should get the vote (once they become indistinguishable from humans). Kurzweil points out that when we build a machine as intelligent as we are, it'll be the last machine we have to build, since the machine itself can create all future machines. This means that if we allow robots to vote, they'll simply overwhelm us by replicating, and then exploiting the power of "one 'man' one vote." Wouldn't this make universal suffrage for all beings (whether bio or techno) impossible in practice, even if philosophically or morally imperative?

(We could e.g. legislate that all beings that are exactly the same get just one vote collectively... but that'd piss off a whole bunch of identical twins!)

Even now, human groups with high birth-rates eventually gain more power at the ballot box. This doesn't create immediate crisis, since humans are so slow and inaccurate in our efforts to reproduce ourselves. When computers can do so without friction -- making their replicas quickly and flawlessly -- it'll be a different matter.


Doug said...

"This doesn't create immediate crisis, since humans are so slow and inaccurate in our efforts to reproduce ourselves. When computers can do so without friction -- making their replicas quickly and flawlessly -- it'll be a different matter."

Take a look at the hard costs here. To raise a child in America, from conception to independent adulthood, costs maybe a few hundred thousand over 20-25 years? That figure includes non-survival aspects of modern life like culture experiences and education, and also includes decades of life lived by the new organism. Reproduction itself costs only the additional food consumed by the mother beyond her own needs while assembling the new copy.

Supercomputers like Watson or Deep Blue, by contrast have hundreds of millions of dollars invested in them, not counting all the supporting research, experimentation and time devoted to programming & teaching them their tasks, and meanwhile they've proven themselves capable of besting humans only in highly specialized tasks; chess and a trivia game respectively.

Fast-forward to the point where machine-life could truly function independently outside of a tightly controlled environment like a factory or a game with artificially constrained rules, and procreation still would not come cheap. Highly refined metals and compounds used in computing today are extremely expensive for a reason: they're hard to come by. Such things wouldn't be any less costly for machines to acquire. And the energy required to live will be just as dear for a machine is it is for humans. Organic chemistry allows life based on little more than dirt, water, and sunshine. Our procreation comes relatively cheap by comparison.

Also, "accurate" reproduction shouldn't necessarily be assumed to be a good or desirable thing. If biological life reproduced itself in perfect copies like a piece of software, it would be vulnerable to perfect extinction. It's the random mutations in biological code that insure against an unforeseen threat wiping out all life at once.

Stuart said...

Hey, Doug, thanks as always for your insight. I'm trying to represent Kurzweil's view a bit here. He'd say that technological advancement is hyper-accelerating (the amount of information that can be stored on a computer chip increasing; the speed of processing this data increasing; the cost of storing and processing data decreasing). The rate of tech advance is double-exponential, since each step (e.g. the spread of the internet and personal computing) becomes a tool that makes the next step come ever quicker.

Meanwhile, biological evolution is so painfully slow. You throw some progeny into the world, and the most fit ones may have a tiny advantage in surviving and procreating and allowing the information stored in DNA to get a little more refined.

So Kurzweil would say that while building a super-computer may be more expensive today vs raising a child, technology will race past biology within a couple decades.

He'd also argue that: intelligent machines on the nano-level will be able to manipulate molecules to produce the raw materials they need to build more.

Whether any of this is good or desirable is a big Don't Know. Kurzweil wants to live forever, and thinks technology will make this possible. He doesn't seem to doubt too deeply why any individual should last forever. I think Buddha would say that everything is changing, changing; everything that appears will eventually dissolve like a floating cloud; so good or bad, like it or not, this "I" is like a flash of lightning or a drop of dew.

Stuart said...

I believe that Kurzweil would say that for the vast vast majority of the history of the universe, advancement/evolution has been entirely trial-and-error (e.g. DNA mutating and the lives thus created living or dying non-randomly).

The human brain made a great leap forward by developing the capacity to preform "what-if" analysis, so we can make advancements so so much faster than trial-and-error would allow. This ability has been greatly enhanced by technology, which will eventually become entirely indistinguishable (or merged with) biology.