3 ms·

There is a difference in the problems that AI might create in the future (destroying all humans) vs the problems it creates now. E.g. AI products had worse STT

by est31 8y ago

There is a difference in the problems that AI might create in the future (destroying all humans) vs the problems it creates now. E.g. AI products had worse STT capabilties for people with a spanish accent or from black backgrounds than american english speakers. Other AIs had more errors with black people. This board's mission is not to prevent an AGI to become self aware and destroy all humans as a natural conclusion of being self aware. Personally I think this scenario is pretty unlikely anyway.

xiphias2 8y ago
I'm more worried about attack vs defence assymetry increasing in the world than a self aware AI destroying it. You just need 1 crazy human and extremely strong assymetrical warfare. If you look at the cases when a human killed lots of people with a truck, and see that in 20 years with cheap self flying drones a crazy person can kill even more people, you can extrapolate this and get to a sad conclusion (at least I did that).
- 0x2AF31DA8 8y ago
  https://qntm.org/asteroids https://qntm.org/asteroids > I believe that there is a threshold of power beyond which nobody can be trusted. Where, in fact, it is impossible for any entity to even theoretically demonstrate the track record of judgement, responsibility and infallibility that would be necessary.
  - xiphias2 8y ago
    This was amazing read, thanks! Also I believe it overestimates the power needed to kill all people. How much time does it take for us to get to the point when an entity has live GPS coordinates of all people living in the world for example? Just 15 years ago this question sounded stupid.
- est31 8y ago
  That's indeed a far more likely scenario. There are plenty of crazy humans around right now. It sure is a big threat, but AI can also help to prevent damage. To stay with your truck example, Anis Amri for example would have killed far more had the truck not had a system that automatically engages brakes when there is an obstacle. Similarly, one could build strong AI systems to shoot down any drone attack. Bioweapons are more dangerous IMO because they are harder to defend from.

by est31 8y ago

xiphias2 8y ago

I'm more worried about attack vs defence assymetry increasing in the world than a self aware AI destroying it. You just need 1 crazy human and extremely strong assymetrical warfare. If you look at the cases when a human killed lots of people with a truck, and see that in 20 years with cheap self flying drones a crazy person can kill even more people, you can extrapolate this and get to a sad conclusion (at least I did that).

0x2AF31DA8 8y ago
https://qntm.org/asteroids https://qntm.org/asteroids > I believe that there is a threshold of power beyond which nobody can be trusted. Where, in fact, it is impossible for any entity to even theoretically demonstrate the track record of judgement, responsibility and infallibility that would be necessary.
- xiphias2 8y ago
  This was amazing read, thanks! Also I believe it overestimates the power needed to kill all people. How much time does it take for us to get to the point when an entity has live GPS coordinates of all people living in the world for example? Just 15 years ago this question sounded stupid.
est31 8y ago
That's indeed a far more likely scenario. There are plenty of crazy humans around right now. It sure is a big threat, but AI can also help to prevent damage. To stay with your truck example, Anis Amri for example would have killed far more had the truck not had a system that automatically engages brakes when there is an obstacle. Similarly, one could build strong AI systems to shoot down any drone attack. Bioweapons are more dangerous IMO because they are harder to defend from.