This feels like it’s the same thing as the low-impact paper.
https://www.dropbox.com/s/cjt5t6ny5gwpcd8/Low_impact_S%2BB.pdf?raw=1
There the AI must maximise U=u−μR where μ>0 is a weight, u is a positive goal, and R is penalty function for impact.
Am I mistaken in thinking that T is the same as u and I is the same as −μR?
It’s not the same (but similar), because my proposal is just about learning a model of impact, and has nothing to do with the agent’s utility function.
You could use the learned impact function, I, to help measure (and penalize) impact, however.
This feels like it’s the same thing as the low-impact paper.
https://www.dropbox.com/s/cjt5t6ny5gwpcd8/Low_impact_S%2BB.pdf?raw=1
There the AI must maximise U=u−μR where μ>0 is a weight, u is a positive goal, and R is penalty function for impact.
Am I mistaken in thinking that T is the same as u and I is the same as −μR?
It’s not the same (but similar), because my proposal is just about learning a model of impact, and has nothing to do with the agent’s utility function.
You could use the learned impact function, I, to help measure (and penalize) impact, however.