Vidal's library
Title: Provably Bounded-Optimal Agents
Author: Stuart J. Russell and Devika Subramanian
Journal: Journal of Artificial Intelligence Research
Volume: 2
Pages: 575--609
Year: 1995
Abstract: Since its inception, artificial intelligence has relied upon a theoretical foundation centered around perfect rationality as the desired property of intelligent systems. We argue, as others have done, that this foundation is inadequate because it imposes fundamentally unsatisfiable requirements. As a result, there has arisen a wide gap between theory and practice in AI, hindering progress in the field. We propose instead a property called bounded optimality. Roughly speaking, an agent is bounded-optimal if its program is a solution to the constrained optimization problem presented by its architecture and the task environment. We show how to construct agents with this property for a simple class of machine architectures in a broad class of real-time environments. We illustrate these results using a simple model of an automated mail sorting facility. We also define a weaker property, asymptotic bounded optimality (ABO), that generalizes the notion of optimality in classical complexity theory. We then construct universal ABO programs, i.e., programs that are ABO no matter what real-time constraints are applied. Universal ABO programs can be used as building blocks for more complex systems. We conclude with a discussion of the prospects for bounded optimality as a theoretical basis for AI, and relate it to similar trends in philosophy, economics, and game theory.

Cited by 143  -  Google Scholar

@Article{russell95a,
  author =	 {Stuart J. Russell and Devika Subramanian},
  title =	 {Provably Bounded-Optimal Agents},
  journal =	 {Journal of Artificial Intelligence Research},
  year =	 1995,
  volume =	 2,
  pages =	 {575--609},
  abstract =	 {Since its inception, artificial intelligence has
                  relied upon a theoretical foundation centered around
                  perfect rationality as the desired property of
                  intelligent systems. We argue, as others have done,
                  that this foundation is inadequate because it
                  imposes fundamentally unsatisfiable requirements. As
                  a result, there has arisen a wide gap between theory
                  and practice in AI, hindering progress in the
                  field. We propose instead a property called bounded
                  optimality. Roughly speaking, an agent is
                  bounded-optimal if its program is a solution to the
                  constrained optimization problem presented by its
                  architecture and the task environment. We show how
                  to construct agents with this property for a simple
                  class of machine architectures in a broad class of
                  real-time environments. We illustrate these results
                  using a simple model of an automated mail sorting
                  facility. We also define a weaker property,
                  asymptotic bounded optimality (ABO), that
                  generalizes the notion of optimality in classical
                  complexity theory. We then construct universal ABO
                  programs, i.e., programs that are ABO no matter what
                  real-time constraints are applied. Universal ABO
                  programs can be used as building blocks for more
                  complex systems. We conclude with a discussion of
                  the prospects for bounded optimality as a
                  theoretical basis for AI, and relate it to similar
                  trends in philosophy, economics, and game theory.},
  keywords =     {learning bounded-rationality ai},
  url = 	 {http://jmvidal.cse.sc.edu/library/russell95a.pdf},
  googleid = 	 {vK6j5ogUhqcJ:scholar.google.com/},
  arxiv = 	 {cs/9505103},
  cluster = 	 {12071358429430787772}
}
Last modified: Wed Mar 9 10:14:03 EST 2011