4 ms·

The AI control problem is a sci-fi narrative that is less instructive as a future speculation, and more instructive as an allegory for the present, in which a v

by aabhay 6y ago

The AI control problem is a sci-fi narrative that is less instructive as a future speculation, and more instructive as an allegory for the present, in which a vanishingly small upper elite power class can construct a false reality for a destitute proletariat.

ALittleLight 6y ago
Why do you think that? My own intuition says that AI Control is a huge problem in the literal sense and not a metaphor at all. My reasoning: 1. No known reason why AGI can't exist. We know our general intelligence exists, so unless there is something magic about us, AGI should be possible. 2. Engineered things can surpass biological things on specific dimensions. A jet flies faster than a bird, a crane lifts more than an elephant, car faster than cheetah, etc. If the same concept holds for intelligence, we could build something much smarter than us. Put another way, we don't know how far above us the limit of intelligence is, so there is some unknown space into which our new intelligence may fall. It could be much smarter than us. 3. Recursive self-improvement, digital copies, horizontal scaling, never getting tired, faster physical limits on computation, perfect memory - AGI may have lots of advantages over us. 4. Intelligence more powerful than our own could do anything, including taking over the world, exterminating humanity, creating a virtual hell and trapping everyone in it. The AGI may not have a motive to do so, but if it is controlled by the wrong people, the people could provide the bad motives. 5. It seems like, however far we are from the singularity, we are getting closer. Our computers continue to do more intelligent things. 6. We don't yet have a good way to control AGI if we create it. In short, it seems like we are approaching something very dangerous and we don't know what to do when we get there. Seems like a literal problem to me.
- athrowaway3z 6y ago
  >No known reason why AGI can't exist >it seems like we are approaching Sure on 1. But there is a jump in complexity required to do what a brain does. I'm not sure what the actual numbers are, but complexity doesn't scale linearly when everything is connected. We only have a single successful example of a super organism growing exponentially in its complexity. ( for now ). That took evolution millions of years from the first eye's doing image recognition distributed over untold number of brains. Us playing AI gods will surely speed it up, but not before something like global warming floods my floors. And where those kind of issues require action now, the issue of AI can be postponed at any moment by a power outage.

by aabhay 6y ago

ALittleLight 6y ago

Why do you think that? My own intuition says that AI Control is a huge problem in the literal sense and not a metaphor at all. My reasoning: 1. No known reason why AGI can't exist. We know our general intelligence exists, so unless there is something magic about us, AGI should be possible. 2. Engineered things can surpass biological things on specific dimensions. A jet flies faster than a bird, a crane lifts more than an elephant, car faster than cheetah, etc. If the same concept holds for intelligence, we could build something much smarter than us. Put another way, we don't know how far above us the limit of intelligence is, so there is some unknown space into which our new intelligence may fall. It could be much smarter than us. 3. Recursive self-improvement, digital copies, horizontal scaling, never getting tired, faster physical limits on computation, perfect memory - AGI may have lots of advantages over us. 4. Intelligence more powerful than our own could do anything, including taking over the world, exterminating humanity, creating a virtual hell and trapping everyone in it. The AGI may not have a motive to do so, but if it is controlled by the wrong people, the people could provide the bad motives. 5. It seems like, however far we are from the singularity, we are getting closer. Our computers continue to do more intelligent things. 6. We don't yet have a good way to control AGI if we create it. In short, it seems like we are approaching something very dangerous and we don't know what to do when we get there. Seems like a literal problem to me.

athrowaway3z 6y ago
>No known reason why AGI can't exist >it seems like we are approaching Sure on 1. But there is a jump in complexity required to do what a brain does. I'm not sure what the actual numbers are, but complexity doesn't scale linearly when everything is connected. We only have a single successful example of a super organism growing exponentially in its complexity. ( for now ). That took evolution millions of years from the first eye's doing image recognition distributed over untold number of brains. Us playing AI gods will surely speed it up, but not before something like global warming floods my floors. And where those kind of issues require action now, the issue of AI can be postponed at any moment by a power outage.