Department of Theoretical Philosophy
Realist and anti-realist perspectives on Solomonoff Prediction.
The 1960 and 1964 papers of Raymond J. Solomonoff mark the birth of the branch of theoretical computer science that is now known as algorithmic information theory or Kolmogorov complexity. The theory of prediction that emerges from those papers can be summarized in the slogan that in a universal setting, there are universal predictors. Here the term `universality' is to informally convey the feature of full generality, of including everything.
The question of the universality of the setting revolves for large part around the status of the central assumption of computability. Different positions regarding this assumption give rise to different interpretations of the formal setting. The aim of this paper is to provide an overview of the formal setting of Solomonoff Prediction and its possible interpretations.
It will prove to be natural to frame this discussion in terms of the scientific realism debate. Indeed, the common and arguably most straightforward interpretation of Solomonoff Prediction can be seen to adopt the two main realist theses. The drawback of this interpretation is its commitment to a questionable metaphysical assumption of computability. By dropping one of the two theses we arrive at two different antirealist interpretations. I argue that the second of these interpretations, that trades the reliability of the universal predictors for their optimality, holds most promise.
|Last modified:||01 November 2013 2.11 p.m.|