My view isn't that extreme. So I presume yours as well isn't as extreme as I seem to think. In fact, upon further inspection we might be closer to agreement than it seems
I'm firstly saying that it's perfectly possible for artificial intelligence not to have any urge for self-preservation. In addition I expect that an artificial intelligence engineered from the ground up will certainly be free from any of the emotional traits or shortcomings that science-fiction writers seem to expect as a matter of fact.
The reason why I think this is that the human brain only has those emotions because they are driven by specialised nerve centres. That this is so is gleaned from observing people who've unfortunately had them knocked out. Emotions are not emergent properties of intelligence. Like naturally occurring intelligence they are only emergent properties of an evolved biological system. So if we don't put emotional processing into intelligent programs, they won't have them.
If artificial intelligence is produced simply by letting it evolve freely all bets are off though. There's no longer a guarantee that the evolved solution won't include some unexpected extras. Designed AI will be free from pesky emotions but evolved AI might well not be.
Personally though I don't give a purely genetic approach to creating AI much chance. The amount of interaction each iteration needs to have would be prohibitively costly if something approaching the degree of specialisation of the various human nerve centres is to be reached without guidance from the results of the billions of years we've had.
Because I expect any reasonably intelligent computer programs to be designed instead of evolved, I don't see any danger. The only danger I see is that we might rely so much on thinking machines that we forget to develop our own intelligence, or more precisely that of our children and grandchildren. An altogether different matter.
Jay Kadis wrote on Mon, 10 January 2011 16:30 |
Just substitute "group" for species if you don't like the idea of clear delineation of species.
|
Indeed, "group selection" is the correct term for the now well discredited interpretation that genes evolve to instigate behaviour benefiting collections of individuals other than kin. Social behaviours extending beyond kin are all analysed -successfully- in terms of game theory (look up "evolutionarily stable strategy").
The concept of "group selection" has mainly found entry in our collective consciousness because of the infamous lemming film that people thought showed lemmings trying to save the species from collapse. This is clearly impossible because lemmings that did not have such a gene would not jump, therefore upping the prevalence of non-jumping alleles. In fact, evolution regularly drives organisms to extinction. A gene that makes individuals more capable of using a limited resource compared to others carrying alleles will become more and more prevalent in an ever shrinking population.
One scenario of group selection that isn't fully ruled out occurs when in some way a population manages to remain extremely homogenous. This is the de facto situation in species like bees where an entire colony share all of their genes but then of course that's just an extreme case of kin selection and not group selection as it's normally understood. Other than that, sufficient homogeneity for group selection to work has never been seen. Group selection is very much part of "folk-evolution-theory" as seen in popular discourse.