People who have only encountered optimization as a qualitative idea and not in a technical context seem to assume “global” can mean “unbounded/infinite domain”.
No you can’t even define optimization problems in such cases without tricks like exponential decay discounting
Conversation
Replying to
This has particular implications for how you think about infinite time horizon problems. If you want to think in optimization terms you’ll be forced to value future and past against present in constrained ways that may not meet intent
1
5
Replying to
I understand that. I was unclear on what particular finite game you were looking to measure “optimized” against, since you didn’t seem entirely clear on that yourself. :)
1
Show replies
Replying to
Can you give me an example of what you mean here I am intrigued but don’t quite grok it
1
Replying to
I think empowerment measures provide a backdoor around this by proxying for prospects in unbounded contexts
arxiv.org/pdf/1310.1863.
1



