I evaluate Schurz's proposed meta-inductive justification of induction, a refinement of Reichenbach's pragmatic justification that rests on results from the machine learning branch of prediction with expert advice.My conclusion is that the argument, suitably explicated, comes remarkably close to its grand aim: an actual justification of induction. This finding, however, is subject to two main qualifications, and still disregards one important challenge.The first qualification concerns the empirical success of induction. Even though, I argue, Schurz's argument does not need to spell out what inductive method actually consists in, it does need to postulate that there is something like the inductive or scientific prediction strategy that has so far beensignificantlymore successful than alternative approaches. The second qualification concerns the difference between having a justification for inductive method and for sticking with inductionfor now. Schurz's argument can only provide the latter. Finally, the remaining challenge concerns the pool of alternative strategies, and the relevant notion of a meta-inductivist's optimality that features in the analytic step of Schurz's argument. Building on the work done here, I will argue in a follow-up paper that the argument needs a strongerdynamicnotion of a meta-inductivist's optimality.

The no-free-lunch theorems promote a skeptical conclusion that all possible machine learning algorithms equally lack justification. But how could this leave room for a learning theory, that shows that some algorithms are better than others? Drawing parallels to the philosophy of induction, we point out that the no-free-lunch results presuppose a conception of learning algorithms as purely data-driven. On this conception, every algorithm must have an inherent inductive bias, that wants justification. We argue that many standard learning algorithms should rather be understood as model-dependent: in each application they also require for input a model, representing a bias. Generic algorithms themselves, they can be given a model-relative justification.

Putnam (1963) construed the aim of Carnap's program of inductive logic as the specification of a "universal learning machine," and presented a diagonal proof against the very possibility of such a thing. Yet the ideas of Solomonoff (1964) and Levin (1970) lead to a mathematical foundation of precisely those aspects of Carnap's program that Putnam took issue with, and in particular, resurrect the notion of a universal mechanical rule for induction. In this paper, I take up the question whether the Solomonoff-Levin proposal is successful in this respect. I expose the general strategy to evade Putnam's argument, leading to a broader discussion of the outer limits of mechanized induction. I argue that this strategy ultimately still succumbs to diagonalization, reinforcing Putnam's impossibility claim.

Abstract. Algorithmic information theory gives an idealized notion of compressibility, that is often presented as an objective measure of simplicity. It is suggested at times that Solomonoff prediction, or algorithmic information theory in a predictive setting, can deliver an argument to justify Occam's razor. This paper explicates the relevant argument, and, by converting it into a Bayesian framework, reveals why it has no such justificatory force. The supposed simplicity concept is better perceived as a specific inductive assumption, the assumption of effectiveness. It is this assumption that is the characterizing element of Solomonoff prediction, and wherein its philosophical interest lies.

This paper poses a challenge to Schurz's proposed meta-inductive justification of induction. It is argued that Schurz's argument requires a notion of optimality that can deal with an expanding pool of prediction strategies.