Decision-making for an Agent with Bounded Resources
C. Manfred Department of Computer Science, University of Illinois at Urbana-Champaign
We consider the behavior of a decision-making agent with bounded resources, which faces problems whose complete solution exceeds those resources. While there are several available methods to treat the resulting “logical uncertainty,” a rigorous treatment has until now escaped realization. We show that, as intuitively expected, there exists an optimal decision-making procedure analogous to expected-utility maximization. We present two practical algorithms for determining the weights in this procedure, which are shown to be correct in the extreme limits. When these algorithms are interpolated, the resulting weights do not differ from the optimum by more than a small constant.
G. Branwen Department of Commenting, LessWrong at The Internets
In Manfred (2012), the author presents a result on an asymptotically ideal utility-maximization of a function in the presence of incomplete information. We discuss the hidden assumption used in the proof, and exhibit a repaired proof which can be shown to be only a special case of long-established asymptotic universal search algorithms based on Solomonoff induction over all computably enumerable functions, from which the Manfred algorithm does not improve except by a constant factor ε in limited domains.
Both abstracts are pretty impressive. I certainly could not tell the difference from a real thing. Maybe this should be a regular feature. Care to fake something EY would author?
As realistic as you can make it. Strive to fool the experts in the area.
Decision-making for an Agent with Bounded Resources
C. Manfred
Department of Computer Science, University of Illinois at Urbana-Champaign
We consider the behavior of a decision-making agent with bounded resources, which faces problems whose complete solution exceeds those resources. While there are several available methods to treat the resulting “logical uncertainty,” a rigorous treatment has until now escaped realization. We show that, as intuitively expected, there exists an optimal decision-making procedure analogous to expected-utility maximization. We present two practical algorithms for determining the weights in this procedure, which are shown to be correct in the extreme limits. When these algorithms are interpolated, the resulting weights do not differ from the optimum by more than a small constant.
Comment on Manfred 2012
G. Branwen
Department of Commenting, LessWrong at The Internets
Both abstracts are pretty impressive. I certainly could not tell the difference from a real thing. Maybe this should be a regular feature. Care to fake something EY would author?