Friday, 12 November 2004
Ockham's Razor and the Highway to the Truth:
A Kinky Resolution of the Realism Debate
Kevin T. Kelly
Carnegie Mellon University
12:05 pm, 817R Cathedral of Learning

Abstract: Here is the crux of the debate concerning scientific realism. When several, alternative theories fit the available data, Ockham's razor enjoins you to choose the simplest. But how could such a policy possibly help you find the truth? For Ockham's razor is a fixed bias toward simplicity that can no more indicate truth than a broken thermometer stuck on a particular reading can indicate temperature. Standard responses either (1) beg the question in favor of simplicity by assuming a prior probabilistic bias toward simple worlds or (2) bait-and-switch by substituting some extraneous aim (e.g., testability, symmetry, predictive accuracy) for finding the true theory.

The linchpin of the puzzle is the tacit premise that Ockham's razor could only help you find the true theory by indicating or pointing at it like a compass needle. But typical advice for how to get somewhere isn't like that: it puts you on the straightest (least kinky) route to your goal without necessarily pointing at the goal. The best such advice may even point opposite to the goal (e.g., if you have to backtrack a few blocks to get on the freeway). I will argue that Ockham's advice works in a very similar way. Indeed, I will prove that Ockham's razor is the unique scientific strategy that minimizes content-losing retractions or scientific revolutions en route to the truth (in a precise, worst-case sense). According to the demonstration, seeking simple theories helps you find complex truths in much the same way that turning away from your goal to get on the freeway helps you get there. So Ockham's razor does help you find the truth in a strong and unique sense, but it doesn't indicate the truth in the present since arbitrarily many severe retractions may still await a retraction-minimizing scientist in the future (the problem of induction). This argument vindicates both the realist's appeals to simplicity and the anti-realist's skeptical doubts. The approach is broadly applicable to statistical model selection, to causal inference, to Goodman's riddle, and even to purely formal problems in computability theory.

The talk is self-contained and is aimed at a general philosophical and scientific audience. Animated diagrams illustrate the several novel concepts involved.