4 ms·
Here is the thing about all of those, none of them give concrete step by step examples of how we would get to "unfriendly" or "friendly" AGI respectively. Now,
by AndrewKemendo 12y ago
Here is the thing about all of those, none of them give concrete step by step examples of how we would get to "unfriendly" or "friendly" AGI respectively. Now, you might say - well of course because we don't know how an AGI will be built yet. Right! That is exactly the point.
At this point all of the MIRI (Former SIAI) staffers have stopped trying to explain paths to unfriendly AGI and have been just assuming them for their solutions. Just look at Yudkowski's work a few years ago on CEV and you'll see how any attempt at formalization falls apart.
Superintelligence just takes from the starting point they seeded in Global Catastrophic Risks, but doesn't actually address it in any more detail IMO. "Our Final Invention" tried also to do this but failed as well.
To be clear the AGI community understands thoroughly the risks inherent in AGI - I would venture to say far more-so than the non-researcher. So they are worth exploring, but at a certain point you have to say "We aren't sure if the atmosphere will catch on fire or not but lets try it anyway because it is the next logical step."
At some point we need to have the hard conversation about what humans will do in a world where we are less relevant.
Or just stop pursuing it altogether and put a moratorium on AGI research. A terrible terrible idea in my opinion.