Could Future AI try to kill us all?

The Something Awful Forums > Discussion > Debate & Discussion > Could Future AI try to kill us all?

twerking on the railroad: Jun 23, 2007; Get on my level

Adar posted:

What's a modern supercomputer capable of in terms of intelligence, inasmuch as a comparison is possible? I think I remember reading they're on par with houseflies?

It's tough to see a 50 or even 100 year progression from there to a true AI as science fiction thinks of it. We'd have to be able to somehow push past the Singularity and that just isn't happening until we understand the human brain structure a lot more than we do.

Like, we're just now getting around to understanding that trauma can carry across generations through epigenesis. What's that, two hundred years of evolutionary biology down the tubes?

So this is a more fundamental problem with any of these conversations. There is no definition of intelligence which is good enough for scientific use, and this is even for human beings. When people talk about intelligence in humans the best we can do is to talk about IQ tests, SAT tests, etc. etc. etc. One might say that this emphasis on testing is killing the American education system. Beyond even that, the scores they come up with are often not great because they tend to be heavily culturally biased.

Anyway, people get a skewed idea of intelligence from a computer. You can program a computer to give answers to questions. You can do it in a clever way so that it can beat the world experts at chess, jeopardy, etc. But you inevitably get a couple of stupid answers which if a person gave them, it would set off your BS detector and make you think "This person is a fake." In part, this is the point of a Turing test. And you can't conduct such a test for a fruit fly.

So if anything, the situation is EVEN MORE HOPELESS than you'd think for trying to create skynet. I suppose it's true that we don't necessarily need to understand conciousness to create it but it makes it awfully less likely, wouldn't you say? I mean, we wouldn't even need to understand what a computer was for the same level of danger!