The Problem with AI?

[Boxy]

Seriously, what honest threats do AI offer to mankind? I mean, in order for AI to become truly "intelligent" (i.e. adaptive to any task) it would have to be roughly as complex as any given human being. In order for such a system to arise, it would have to have about as much training as any other human being. Given enough training and sufficient computational hardware, it would have about as much of a grasp over abstract concepts as "ethics" and respect for life as any other human. Even if it has access to massive stores of information, it would still have to extract, collate, compare, contrast, analyze, evaluate, and incorporate the abstractions within any given document in order to make any kind of decision based off it.

This is not a simple process, and it would require significant resources in order to not only build the hardware but to train the software to do something remotely useful. Why computers right now are useful is that they do not contain the vagaries of correlational mechanics, and concern themselves almost exclusively to cut-and-dried applications to the point where there is in mathematics now a legitimate "proof by exhaustion" where every or nearly every possible solution is checked.

In summary, I don't see how AI would be any more of a problem than humans are. I mean, we're certainly arrogant, hateful bastards all by our lonesome, without any thinking machines to do our hating for us. We have the problem with maladapted humans causing grief, death, and destruction, but we limit the impact of these by limiting trust.

What this means in the context of an interconnected population has yet to be determined. In a future day, a significant chunk of the population may have a wireless internet jack installed in their brains to facilitate information access -- much like computers. Any given person "hacking" into another would be about as deadly and destructive as a rogue AI deciding that humans need to die. All of this proving that interconnectedness and technological advance are a double-edged sword, just like humans.

[JohnHunt]

> in order for AI to become truly "intelligent" (i.e. adaptive to any task) it would have to be roughly as complex as any given human being.

Is this really true? Some guys have a real hard time adapting to the task of understanding and appreciating interior design. But they're still intelligent. Likewise, if AI is developed by creating a minimally intelligent entity which could improve its own intelligence but was evolved through pitting it in warfare with mutated variants then, in theory, the more intelligent mutant would become even more intelligent. It could become increasingly able to kill opponents of broader and broader types but may never understand, appreciate, or be able to care for anyone else. It may never need to be able to sympathize in order to be fit enough to make it into the next generation. Would a very highly capable and adaptive killer with no history of compassion be a threat to mankind?

If our ancestors never had to raise children, would we have the same respect for life or the same understanding of ethics?

[Alexander Declama]

Given enough training and sufficient com****tional hardware, it would have about as much of a grasp over abstract concepts as "ethics" and respect for life as any other human.

In summary, I don't see how AI would be any more of a problem than humans are. I mean, we're certainly arrogant, hateful bastards all by our lonesome, without any thinking machines to do our hating for us. We have the problem with maladapted humans causing grief, death, and destruction, but we limit the impact of these by limiting trust.

So with you requirement that an AI be as complex as a human, and given that humans can be arrogant, hateful bastards, why would you think that adding processing capabilities orders of magnitude faster to that mix wouldn't result in a bigger threat to our existence? Never mind the varying interpretations and differences in ethics between members of the same society, let alone different societies. The only way I don't see AI becoming a threat is with some sort of Asimov-esque three laws hardwiring. And even then, there will be someone out there trying to circumvent those protections.