First? Swing low, see how it performs, especially with a long-term project. Something low-stakes. Maybe something like a populated immersive game world. See what comes from there. Is it stable? Is it sane? Does it keep to its original parameters? What are the costs of running the agent/system? Can it solve social alignment problems?
Heck, test out some theories for some of your other answers in there.
Thank you for the comment. I think all of what you said is reasonable. I see now that I probably should’ve been more precise in defining my assumptions, as I would put much of what you said under “…done significant sandbox testing before you let it loose.”
I kind of think of this as more than sandbox testing. There is a big difference between how a system works in laboratory conditions, and how it works when encountering the real world. There are always things that we can’t foresee. As a software engineer, I have seen system that work perfectly fine in testing, but once you add a million users, then the wheels start to fall off.
I expect that AI agents will be similar. As a result, I think that it would be important to start small. Unintended consequences are the default. I would much rather have an AGI system try to solve small local problems before moving on to bigger ones that are harder to accomplish. Maybe find a way to address the affordable housing problem here. If it does well, then consider scaling up.
First? Swing low, see how it performs, especially with a long-term project. Something low-stakes. Maybe something like a populated immersive game world. See what comes from there. Is it stable? Is it sane? Does it keep to its original parameters? What are the costs of running the agent/system? Can it solve social alignment problems?
Heck, test out some theories for some of your other answers in there.
Thank you for the comment. I think all of what you said is reasonable. I see now that I probably should’ve been more precise in defining my assumptions, as I would put much of what you said under “…done significant sandbox testing before you let it loose.”
I kind of think of this as more than sandbox testing. There is a big difference between how a system works in laboratory conditions, and how it works when encountering the real world. There are always things that we can’t foresee. As a software engineer, I have seen system that work perfectly fine in testing, but once you add a million users, then the wheels start to fall off.
I expect that AI agents will be similar. As a result, I think that it would be important to start small. Unintended consequences are the default. I would much rather have an AGI system try to solve small local problems before moving on to bigger ones that are harder to accomplish. Maybe find a way to address the affordable housing problem here. If it does well, then consider scaling up.