![Sample our Humanities journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5939/TOC-Subject-Sample-2021-Humanities.jpg"})
ABSTRACT
Dennett [1991] proposes a novel ontological account of the propositional attitudes—real patterns. Despite its name, the degree to which this account is committed to realism remains unclear. In this paper, I propose an alternative criterion of pattern instantiation, one that assesses the difficultly of faithfully interpreting a physical system as instantiating a particular pattern. Drawing on formal measures of simplicity and similarity, I argue that, for well-instantiated patterns, our interpretation will be computable by using a short program. This approach preserves the flexibility of Dennett’s original, while offering substantially stronger realist commitments.
Disclosure Statement
No potential conflict of interest was reported by the author.
Notes
1 There might be more than one shortest program, but I have retained the definite article to avoid the awkwardness of using the indefinite article with a superlative in English.
2 There are analogous results in Shannon information theory [Shannon Citation1948], but the algorithmic approach is more natural here since programs make explicit the patterns exploited in compression. For an overview of the relationship between Shannon information and Kolmogorov complexity, see Grünwald and Vitányi [Citation2004].
3 The choice of encoding scheme and reference universal Turing machine complicates matters, but there are convergence guarantees for different choices that suffice for practical purposes [Sipser Citation2013]. See Millhouse [Citation2019] for a philosophical discussion of reference machine selection.
4 The idea that modelling can be viewed as a kind of compression has also been proposed in data science/information theory [C. Wallace Citation2005; Rathmanner and Hutter Citation2011]. However, I argue [Citation2021] that these proposals differ in important ways from Dennett’s.
5 Given a richer set of variables (e.g. one that includes processing time), we might be able to distinguish between these algorithms via predictive economy. However, Dennett gives no account of when a data set is sufficiently rich to support ontological conclusions about a model’s posits, and the information-theoretic justification for real patterns does not (by my lights) suggest one.
6 To be clear, nothing about Dennett’s view precludes our applying his criterion to a model whose data are computational state histories rather than input/output pairs. In this case, a simple and predictive model would indicate a particular pattern in those states. However, in this case, the computational states are not an explanatory posit of the model; they are (presumably) a pre-established feature of the target system.
7 A discussion of the accuracy conditions for fundamental physical models would require a separate treatment. Fortunately, even if we assume an accurate low-level model of a system, deep questions remain about the instantiation of higher-level models.
8 In some cases, we might be interested in assessing only some aspects of our model (e.g. what it says about particular high-level variables). To do so, we might consider a mapping between the states of a physical system and a proper subset of the variables employed by our model.
9 I am indebted to Shaun Nichols, Daniel Dennett, Jenann Ismael, Richard Healey, David Wallace, Kathleen Creel, Brandon Ashby, Nathaniel Oakes, Jeri Millhouse, and Amanda Romaine for their comments, advice, and encouragement. Any remaining errors or omissions are entirely my own.