AGIRI Workshop Panel on Friendly AI Thursday, Sep 7 2006
friendly ai 9:13 pm
Contains both seriousness and a decent amount of humor. Interesting to see people working on AGI discussing Friendly AI in the same room for almost the first time ever. Hopefully this will start happening every year. Panelists: Ari Heljakka, Jeff Medina, Eliezer Yudkowsky, Dr. Karl H. Pribram, and Dr. Hugo de Garis. There is a good diversity of opinion and each person represents a different point of view towards FAI.
Most encouraging of all I think is to see people talking about producing forms of general artificial intelligence, not the very specific (expert) applications which have been the usual subject of AI conferences for a long time.
As Ben Goertzel said in one of his videos, over five years ago if you went to some AI conference and started talking about producing “general intelligence” the research establishment would have just thrown popcorn at you (or worse), but now the idea is treated more seriously.
You should join the SL4 and AGI lists. We’ve been talking about this stuff on a technical and serious level for years. Most of the people in the room are members of these lists.
Karl H. Pribram: Creating a child is easy and fun.
Eliezer Yudkowsky: That’s like saying creating a bottle of coke is easy because you just need to put quarters into the vending machine and out pops the bottle of coke.
…so where are the female transhumanists?
I wonder what this discussion would have been like without anthropomorphism. That is, if all the the participants were thinking of AIs as programs rather than implicitly thinking of them as weird kinds of people.
The latter error is very natural: examples include most reasoning by analogy with humans, and imagining AI behaviour by thinking of what you might do if you were the AI.
The above reasoning method, anthropomorphism, is an example of a heuristic. This works pretty well when we are reasoning about other humans. However, it leads to the bias of imagining that AIs necessarily behave like humans.
This is an example of what the heuristics and biases field studies. Correcting these sorts of error is one reason to study this field, as Eliezer Yudkowsky mentioned.