Rerum Concordia Discors: Robustness and Discordant Multimodal Evidence

Abstract

Rain today, I reckon, given the grey clouds above, the falling barometer, and after all, it is an autumn day in London. My conjecture is supported with multimodal evidence: the clouds, the barometer, the season. The term “multimodal evidence” will be unfamiliar to most, though it is a common intuition that multimodal evidence is valuable. A “mode” is a way of finding out about the world; a type of evidence; a technique or study design. We usually have evidence for or against a hypothesis which comes from a variety of different modes; I call this multimodal evidence. For example, when devising his laws of motion, Newton had evidence on the orbits of the moons of Jupiter and Saturn, the patterns of spring and neap tides at the solstice and equinox, and terrestrial dynamics. Robustness – the state in which hypotheses are supported with concordant multimodal evidence – is one way in which the value of multimodal evidence has been explicated (Section 9.2). Another way in which multimodal evidence is said to be valuable is based on the notion of security (Section 9.3). An empirical challenge for robustness is that when multimodal evidence is available for a particular hypothesis, it is often discordant (Section 9.4) – discordance is ubiquitous. A conceptual challenge is to know when and how modes are sufficiently independent to count as providing multimodal evidence (Section 9.5). A methodological challenge is that to know the impact multimodal evidence should have on our belief in a hypothesis, the multimodal evidence must be assessed and amalgamated by an amalgamation function (Section 9.6). I argue that an amalgamation function for multimodal evidence should do the following: evidence from multiple modes should be assessed on prior criteria (quality of mode), relative criteria (relevance of mode to a given hypothesis) and posterior criteria (salience of evidence from particular modes and concordance/discordance of evidence between modes); the assessed evidence should be amalgamated; and the output of the function should be a constraint on our justified credence. Without principled methods of amalgamating multimodal evidence, appeals to multimodal evidence are vague and inconclusive. Such amalgamation functions could provide more rigorous guidance for our belief in a hypothesis when presented with multimodal evidence.