- Shopping Bag ( 0 items )
This book presents sequential decision theory from a novel algorithmic information theory perspective. While the former is suited for active agents in known environments, the latter is suited for passive prediction in unknown environments. The book introduces these two different ideas and removes the limitations by unifying them to one parameter-free theory of an optimal reinforcement learning agent embedded in an unknown environment. Most AI problems can easily be formulated within this theory, reducing the conceptual problems to pure computational ones. Considered problem classes include sequence prediction, strategic games, function minimization, reinforcement and supervised learning. The discussion includes formal definitions of intelligence order relations, the horizon problem and relations to other approaches. One intention of this book is to excite a broader AI audience about abstract algorithmic information theory concepts, and conversely to inform theorists about exciting applications to AI.
Short Tour Through the Book.- Simplicity & Uncertainty.- Universal Sequence Prediction.- Agents in Known Probabilistics Environments.- The Universal Algorithmic Agent AIXI.- Important Environmental Classes.- Computational Aspects.- Discussion.
Posted March 3, 2005
Solomonoff's famous inference model solves the inductive learning problem in a universal and provably very powerful way. Many methods from statistics (maximum likelihood, maximum entropy, minimum description length...) can be shown to be special cases of the model described by Solomonoff. However Solomonoff Induction has two significant shortcomings: Firstly it is not computable, and secondly it only deals with passive environments. Although many problems can be formulated in terms of sequence prediction (for example categorisation), in AI in general an agent must be able to deal with an active environment where the agent's decisions affect the future state of the environment. In essence, the AIXI model, the main topic of this book, is an extension of Solomonoff Induction to this much more general space of active environments. Although the model itself is very simple (it is really just Solomonoff's model with an expectimax tree added to examine the potential consequences of the agent's actions) the resulting analysis is now more difficult than in the passive case. While optimality can be show in certain senses, the powerful convergence bounds that Solomonoff induction has now appear to be difficult to establish. Like Solomonoff induction, AIXI also suffers from computability problems. In the one of the final sections a modified version of AIXI is presented which is shown to be computable and optimal in some sense. Practically this algorithm would be much too slow, but it is a clear step away from abstract models towards one which can in theory be implemented. For anybody interested in universal theories of artificial intelligence this book is a must. The presentation is quite technical in places and thus the reader should have some understanding of theoretical computer science, statistics and Kolmogorov complexity.Was this review helpful? Yes NoThank you for your feedback. Report this reviewThank you, this review has been flagged.
Posted October 28, 2004
Have you ever wondered if Artificial Intelligence is possible at all? And if so, can it be achieved by a neat and simple construction, rather than some highly complex system that nobody could understand? This book proposes a theory answering both questions in the affirmative. The idea is easy: use a universal model class, treat observations in a probabilistic (Bayesian) way, and do an optimal far-sighted decision on this basis. In the book, this construction is mathematically elaborated, and many properties are stated and proven, such as asymptotic behavior, optimality, etc. On the other hand, no efficient implementation is possible at this stage, since the system requires infinite computation time. This is the only mathematical definition of universal AI I know of, and maybe the only one possible. How can it be implemented efficiently? Does the resulting system behave truly intelligent? Hopefully the future will answer these questions - maybe with your contribution!Was this review helpful? Yes NoThank you for your feedback. Report this reviewThank you, this review has been flagged.