Special Issue ""Algorithmic Complexity in Physics & Embedded Artificial Intelligences"—In Memoriam Ray Solomonoff (1926-2009)"

Assistant Editor
Ms. Laura Simon
MDPI, Kandererstrasse 25, CH-4057 Basel, Switzerland
E-Mail:

Guest Editor
Prof. Dr. Juergen Schmidhuber
IDSIA, Galleria 2, 6928 Manno-Lugano, Switzerland
Website: http://www.idsia.ch/~juergen/
E-Mail:
Interests: artificial intelligence; machine learning; neural networks; Kolmogorow-complexity; robotics

Dear Colleagues,

Is the universe computable, as suggested in the 1940s by Konrad Zuse, inventor of the first working program-controlled computer? With the ascent of virtual realities the idea has become popular, and is now also being taken seriously by physicists, for lack of contrarian physical evidence. Questions to be addressed in this special issue include: Which kind of programs running on which type of computational device could in principle provide a concise description of quantum physics? How can algorithmic complexity theory and Kolmogorov complexity theory guide the quest for simple explanations of the world in the sense of Occam's razor? How do Gödelian limits of mathematics and computation as well as insights from algorithmic information theory restrict the set of valid physical theories, including many world theories? Which sets of computable probability distributions or measures on possible universe histories make sense at all from the perspective of constructive mathematics? Following Solomonoff's theory of optimal inductive inference and algorithmic probability, how can the restrictions embodied by such sets help to predict future events, given past observations in a particular universe? Which testable predictions are made by algorithmic complexity-based theories of physics? Can we in principle design rational decision-making agents or artificial intelligences embedded in computable physics such that their decisions are optimal in reasonable mathematical senses? Which are the fundamental limitations of such decision makers? If physics is hard to compute, can this help to improve cryptography?

Special Issue "In Memoriam Ray Solomonoff" (1926-2009):

The Great Ray Solomonoff, pioneer of Machine Learning, founder of Algorithmic Probability Theory, father of the Universal Probability Distribution, creator of the Universal Theory of Inductive Inference, passed away on Monday 7 December 2009 at age 83. Ray Solomonoff was the first to describe the fundamental concept of Algorithmic Information or Kolmogorov Complexity. In the new millennium his work became the foundation of the first mathematical theory of Optimal Universal Artificial Intelligence. With great sadness the special issue will be "In Memoriam Ray Solomonoff".

Prof. Dr. Juergen Schmidhuber
Guest Editor

Submission

All papers should be submitted to algorithms@mdpi.org. To be published continuously until the deadline and papers will be listed together at the special issue website.

Submitted papers should not have been published nor be under consideration for publication elsewhere. All papers are refereed through a peer-review process. A guide for authors is available on the Instructions for Authors page. Algorithms is an international peer-reviewed quarterly journal published by Molecular Diversity Preservation International.

Article Processing Charges (APC) will be waived for well prepared manuscripts of invited papers. For the first three volumes of this new journal the APC are of 300 CHF (or 550 CHF per paper for those papers that require extensive additional formatting and/or English corrections) for papers submitted before 31 December 2010.

algorithmic complexity
algorithmic information theory
Kolmogorov complexity
descriptive complexity
Kolmogorov-Chaitin complexity
stochastic complexity
algorithmic entropy
program-size complexity
Chaitin entropy
Chaitin complexity

Title: MML, Occam's razor, explanation, prediction, entropy, time's arrow, intelligent design and miracles
Author: David Dowe; E-mail: david.dowe@infotech.monash.edu.au
Abstract: We introduce Minimum Message Length (MML) (Wallace & Boulton, 1968) as a statistically invariant and statistically consistent Bayesian interpretation of algorithmic information theory and Occam's razor. Despite problems of other approaches with retaining statistical consistency under model misspecification (Grunwald & Langford, 2007), no known evidence exists yet of a problem for MML here. We explore the frequently misunderstood distinction between inference (or induction, or explanation) and prediction. We then focus on Wallace's idea (2005) that, given CPT-invariance in Physics, it appears that - contrary to wide-held beliefs - entropy is not time's arrow. We then address why it is that people wish to predict the future but (only) infer (or explain) the past. We draw attention to MML's ability to predict better than rival methods while using simpler models, mentioning some areas of application in physical models. We offer some scepticism about quantum computing, and we use MML to address evidence of "intelligent design'' (Dowe et al., 2009) and miracles (Dowe, 2008a).