Introduction Learning from models Problems with the pretense view Saving the fiction view References Models, Fictions, and Representing Scientific Practice: (Or, I don’t know much about models... but I know that we learn from them*) Corey Dethier University of Notre Dame Philosophy Department corey.dethier@gmail.com March 17, 2018 Models and Simulations *subtitle with apologies to Steven French
Introduction Learning from models Problems with the pretense view Saving the fiction view References Introduction
Introduction Learning from models Problems with the pretense view Saving the fiction view References The fiction view of models Fiction view of models: ‚ Family of positions defended by Barberousse and Ludwig (2009); Frigg (2010a,b); Frigg and Nguyen (2016); Godfrey-Smith (2006); Gr¨ une-Yanoff (2009, 94); Levy (2015); Salis (2016); and Toon (2012). ‚ Models (and their constituents) should be understood or analyzed in the same manner that works of fiction (and their constituents) are analyzed.
Introduction Learning from models Problems with the pretense view Saving the fiction view References The pretense view of models In practice, the fiction view of models is really the pretense view of models. ‚ Defenders of the fiction view analyze models according to Walton (1990)’s “pretense” account of fictions. ‚ Accompanied by a technical move of attaching an operator to the sentences that are “pretend-true”—e.g., “According to the fiction, P ” or “It is true within the pretense that P .”
Introduction Learning from models Problems with the pretense view Saving the fiction view References My argument I’ll argue that: (a) There is a class of comparisons that play an essential role in our ability to learn from models. (b) The technical move introduced above is incapable of rendering these comparisons true; and “true according to the model” won’t cut it. (c) There are technical resources within the literature on fiction that can render such comparisons true. (d) The apparent costs of these resources should be accepted (by defenders of the fiction view).
Introduction Learning from models Problems with the pretense view Saving the fiction view References Learning from models
Introduction Learning from models Problems with the pretense view Saving the fiction view References Learning from models (a first pass) We learn from a model if (and only if?): the model allows us to justify a belief that we were not previously justified in holding. E.g., from a point-mass model of the solar system, we can learn how the planets will accelerate (to a high degree of approximation); we can read the accelerations off of the model. But notice that we can’t learn about the density of the planets from such a model even though we can read these off the model too; the model doesn’t justify conclusions about density.
Introduction Learning from models Problems with the pretense view Saving the fiction view References Learning from models and justification The problem: what distinguishes these cases? Prior knowledge about the model: we know that it accurately represents distances and masses, but misrepresents planet sizes. And the latter plays a role in reaching conclusions about density, while only the former play a role in reaching an (approximate) conclusion about acceleration.
Introduction Learning from models Problems with the pretense view Saving the fiction view References Representing learning from models In order to faithfully represent learning from models, our account of scientific models must allow us to capture the difference between these two cases—between the conclusions that the model justifies and those it doesn’t. Since (in at least some cases), this difference is grounded in similarities and dissimilarities between the model and the target system, faithfully representing learning from models requires that our account allow us to capture these facts about the relationship between model and world.
Introduction Learning from models Problems with the pretense view Saving the fiction view References The necessity of true comparisons In other words, our account of scientific models must allow us to render certain comparisons—such as (1) below— true , or at least to discriminate between them and comparisons like (2). (1) The planets (in the model) have the same mass as they do in the world. (2) The planets (in the model) have the same volume as they do in the world. This distinction between true and false comparisons underwrites learning from models; if we can’t rule some comparisons acceptable in a way that others aren’t, we can’t represent why some conclusions are justified and others aren’t. (Notice that (1) is a comparison between an object in the model and one in the world; this will be important later.)
Introduction Learning from models Problems with the pretense view Saving the fiction view References Problems with the pretense view
Introduction Learning from models Problems with the pretense view Saving the fiction view References The pretense view The pretense view: Models (and their constituents) should be understood as reasoning that occurs within a “pretense.” Effectively, this means attaching a sentence-level operator or distinguishing between true sentences and sentences that are merely “true-in-the-pretense.” E.g., “According to the pretense / model, P .” “It is true-in-the-pretense / -model that P .
Introduction Learning from models Problems with the pretense view Saving the fiction view References Advantages of the pretense / fiction view Does relatively well at representing modeling practice (` a la French and Ladyman [1999]). Like works of fiction, models ‚ come in many shapes and sizes. ‚ are used for many purposes. ‚ have parts that are accurate to the world and parts that aren’t. ‚ teach us lessons about the real world. ‚ require creativity. ‚ employ props or proxies. ‚ facilitate understanding.
Introduction Learning from models Problems with the pretense view Saving the fiction view References How should the pretense view handle (1)? Only two options on the pretense view for capturing what makes (1) a good comparison: Option 1: (1) is true ( tout court ) and (2) is false. Option 2: (1) is false but true-in-the-pretense, while (2) is just false. Neither option will work. First option won’t work: whether or model is abstract or physical and made of foam, etc., the planets just don’t have the same mass in the model and in the world!
Introduction Learning from models Problems with the pretense view Saving the fiction view References What about true-in-the-pretense? Simply: “it is true-in-the-pretense that P” just doesn’t say the same thing as “P.” For example, (3) and (4) have different truth values: (3) Sherlock Holmes is a better detective than any real detective. (4) According to the pretense, Sherlock Holmes is a better detective than any real detective.
Introduction Learning from models Problems with the pretense view Saving the fiction view References Pretense view cannot account for learning Worse: in the context of models, the “according to the pretense” claim just won’t do for the purposes of learning. If (1) is true, then the model is accurate (with respect to mass); if it is accurate, we can learn from it. If (1) is merely true-in-the-pretense, then all we know is that the model says (of itself) that is accurate. That doesn’t license inferences!
Introduction Learning from models Problems with the pretense view Saving the fiction view References Options for the pretense view? Frigg (2010b) suggests a paraphrase strategy: (5) Jupiter (in the model) has the same mass that Jupiter actually has. (6) According to the model, Jupiter has a mass of 1 . 9 ˆ 10 27 kg. Jupiter has a mass of 1 . 9 ˆ 10 27 kg. Having a mass of 1 . 9 ˆ 10 27 kg is equivalent ot having a mass of 1 . 9 ˆ 10 27 kg. I.e., remove all reference to the constituents and only compare the properties.
Introduction Learning from models Problems with the pretense view Saving the fiction view References The necessity of constituents Compare (6) to (7): (7) According to the model, Mercury has a mass of 1 . 9 ˆ 10 27 kg. Jupiter has a mass of 1 . 9 ˆ 10 27 kg. Having a mass of 1 . 9 ˆ 10 27 kg is equivalent ot having a mass of 1 . 9 ˆ 10 27 kg. Why does (6), but not (7), license inferences about the actual solar system? Hard to see how we’re going to spell this out without sneaking in reference to the constituents of the model; after all, what’s wrong is that the model assigns the right mass, but to the wrong object!
Introduction Learning from models Problems with the pretense view Saving the fiction view References Saving the fiction view
Introduction Learning from models Problems with the pretense view Saving the fiction view References A different technical move, part I In order to illustrate my alternative, let’s return to fiction. Compare the following two sentences. (8) According to Leslie Marmon Silko’s Ceremony , Tayo is troubled by his experiences during the war. (9) Tayo, as he is depicted in Leslie Marmon Silko’s Ceremony , is troubled by his experiences during the war. Pre-theoretically, these communicate essentially the same thing. Structurally, however, they’re quite different.
Recommend
More recommend
