Saturday, April 2, 2016

Knowledge, Persons, and Explanation

In my previous post, I offered a clearer way of delineating different theories of explanation. I suggested that we treat epistemic theories of explanation as a subset of pragmatic theories of explanation. Here are the relevant claims:
A theory of explanation is epistemic iffdf according to T, there exists a statement of the form x explains why y” that is true relative to a knowledge corpus K1 and not true relative to another knowledge corpus K2. 
A theory of explanation is pragmatic iffdf according to T, there exists a statement of the form x explains whyy” that is true relative to a person S1 and not true relative to another person S2. 
Stipulation: If a statement is relative to a knowledge corpus, then it is relative to a person.
As I noted, this stipulation requires defense. Here is an initial motivation: knowledge is frequently predicated of people. For instance, John knows that it is raining. This, of course, is also how a majority of professional epistemologists conceive of knowledge. Hence, a knowledge corpus will refer to a/the set of propositions known by an agent.

To be sure, knowledge is sometimes conceived more abstractly, e.g. as a set of propositions rather than a set of states predicated of persons. However, this conception is unstable: it will either collapse into a kind of ontic theory or a kind of pragmatic theory. Let's illustrate this with the idea that knowledge is a justified true proposition. Either this justification is propositional, i.e. it refers to a relation between propositions (p justifies q); or it is doxastic, i.e. it refers to whether a belief is justified (S is justified in believing that q). (This is a common distinction in the epistemological literature.) If justification is propositional, then epistemic theories collapse into a kind of ontic theory. While traditional ontic theories invoke more concrete entities, such as events, mechanisms, causes, etc., epistemic theories would invoke abstract entities such as propositions and whatever relations realize the "propositional justification role." However, to my knowledge, traditional ontic theories have never banned abstract properties and entities.

So, what then, if we think of justification as doxastic? This will lead to my desired result: knowledge becomes characteristic of a person. After all, it's S's belief that is justified, and S is a person.

Monday, March 21, 2016

Ontic, Epistemic, and Pragmatic Models of Explanation

Not so long ago, there were thought to be three main “schools” of explanation in the philosophy of science literature: the ontic, the epistemic, and the pragmatic, e.g. in Wesley Salmon’s masterful “Four Decades of Scientific Explanation.” This tripartite distinction was not employed consistently, and might never have been articulated very precisely. I offer the following as a useful distinction.

A theory T of explanation is ontic iffdf according to T, there exists no statement of the form x explains why y” that is true relative to a person S1 and not true relative to another person S2.

A theory T of explanation is epistemic iffdf according to T, there exists a statement of the form x explains why y” that is true relative to a knowledge corpus K1 and not true relative to another knowledge corpus K2.

A theory T of explanation is pragmatic iffdf according to T, there exists a statement of the form x explains why y” that is true relative to a person S1 and not true relative to another person S2.

Stipulation: If a statement is relative to a knowledge corpus, then it is relative to a person.


According to our stipulation and definitions, all epistemic theories of explanation are pragmatic theories of explanation. Hence, the fundamental divide is between ontic and pragmatic theories. Indeed, it might be more fruitful to describe the distinction as one between "impersonal" and "personal" theories.

More will need to be said about: (a) the viability of our stipulation, (b) the “relativity-clauses” in epistemic and pragmatic models, and (c) what “persons” are in the latter.

Sunday, January 20, 2013

What is a Musical Performance?


Introduction

So, I’ve just begun to read Joe MorrisPerpetual Frontier: The Properties of Free Music, a summary of which can be found here. While I’m still in the early stages, it’s got me thinking about how to organize a group of musicians to play free music, and I’m also seeing how there’s an underlying philosophy that can be useful in communicating some of Morris’ ideas. If I’m feeling up to it, I’ll provide commentary on Morris, perhaps on a per chapter basis. However, mostly I want to hash out my own ideas.
            Suppose we begin with a group of musicians, and our objective is to get them to play “free music.” To make this as concrete as possible, let us suppose that in a month’s time, they will have to play a concert of “free music.” Moreover, suppose that all of the musicians in question are well versed in at least one genre of non-free music, but are relative neophytes in all forms of free music. How would we get these musicians to succeed in their concert?
            It’s tempting to start devising lots of exercises, a la Tom Hall. I can’t stress how important those exercises are, and how much I’ve loved going through them. However, how would we know that these exercises have been effective in producing a successful concert? What we need is a better grasp of the concept of a successful free music performance. An exercise would be effective just in case it promoted that.
I think that philosophy can help on this front. In what I hope will be a series of posts, I’ll show how we can get a clearer concept of a successful free music performance. Today, I’ll focus just on musical performances, temporarily bracketing the success or freedom of such performances.
I suggest that we think of musical performances as systems constituted by mechanisms. Let’s first understand systems and mechanisms more generally (§1), and then discuss how a musical performance is a special case of systems and mechanisms (§2). Third, I’ll consider objections to this view, and offer my replies to them (§3). Finally, I’ll briefly summarize and highlight the practical payoff of this view (§4).

1.   Systems and mechanisms

My goal for today is to show that musical performances can be regarded as mechanistic systems. Before talking about musical performances, let me describe some general features of mechanistic systems. Philosophers and scientists in the relevant disciplines have described many biological, psychological, technological, social, and cultural phenomena as mechanistic systems. While they have different views about what constitutes a mechanism, the general idea about mechanistic systems is this:
(General Idea) Many systems exhibit certain properties because of the properties, activities, and organization of the systems’ parts.

            For now, I’ll rely on your native intuitions as to what constitutes a part-whole relationship, property, activity, and organization. However, I do want to spend a bit more time on a seemingly banal word: the use of “because” in the General Idea, as mechanistic systems have a special “because”-structure.  In his book, Explaining the Brain, (which looks at mechanisms in neuroscience), Carl Craver discusses this “because”-structure, and dubs it the Mutual Manipulability Condition (MMC).
            That may sound scary, but the idea is pretty straightforward: in mechanistic systems, two kinds of changes must be possible:
(1)  Changes to the properties, actions, or organization of the parts must be capable of changing the properties of the system, and
(2)  Changes to the properties of the system (or whole) must be capable of changing the properties, actions, or organization of its parts.
For instance, think about the neuroscience of reading. Here the system is a person reading and the parts are the relevant parts of the brain. The MMC states that:
(1)  Had the relevant brain regions been different, then a person’s way of reading would have been different, and
(2)   Had a person’s way of reading been different, then the relevant brain regions would have been different.
An example of the first manipulation or intervention is brain surgery or the use of certain mind-altering drugs that affect reading. An example of the second would be altering the speed at which words come across a screen in order to change how neurons fire.

2.   Musical performance as a mechanistic system

Now, all of this seems pretty far removed from music, right? Wrong! I'd like to suggest that musical performances are usefully thought of as mechanistic systems. Recall that the ultimate goal is to specify what is constitutive of a successful free music performance. However, for today, I want to make the argument that all musical performances (successful or otherwise, free or otherwise) are mechanistic systems.
            If I’m going to establish that claim, I need to show that: (§2.1) a musical performance is a system composed of parts and (§2.2) the system and parts satisfy the Mutual Manipulability Condition.

2.1.        Musical performances are systems composed of parts

I first want to show that musical performances is an application of the General Idea. I propose that we simply think of a musical performance as a system.
Now clearly, such a performance has many potential properties it can assume. Call these performance properties. Some performance properties are emotive, e.g. performances can be uplifting, dreadful, banal, intense, soothing, etc. Performances properties can also be of a more formal variety, e.g. performances can be in a particular time signature, key, exhibit particular chord progressions, melodies, harmonies, dynamics, etc. (More on performance properties at a later date.)
Moreover, performances are composed of parts. I suggest that we take performers as our primary parts. So a musical performance is a mechanistic system just in case the following is true:
A musical performance exhibits certain properties (e.g. being uplifting, banal intense, etc.) because of the properties, activities, and organization of the performers.
This seems eminently plausible. Moreover, it’s a good thing to impress upon our hypothetical musicians with their hypothetical deadline, for it stresses the responsibility that they have for the musical performance.

2.2.        Musical performances and performers are mutually manipulable

However, if I really want to make my case for musical performances as mechanistic systems, I need to show that musical performances satisfy the MMC. Now, one requirement of the mutual manipulability condition seems relatively uncontroversial, namely this:
(1)  Had the performers’ properties, actions, and organization been different, then the musical performance would have exhibited different properties.
The most obvious evidence for this is that performers play differently to create different musical pieces. However, the other requirement of the mutual manipulability condition is trickier:
(2)  Had the musical performance exhibited different properties, then the performers’ properties, actions, and organization would have been different.
For those who are interested in the puzzle that arises with (2), go to the Appendix. For the rest of you, let me skip straight to its solution: sometimes, one subset of actions, properties, and/or organization of some performers can change another subset of such actions, properties, and/or organization only via the performance properties.
For instance, imagine that all of the members of a group are playing the same rhythmical motif, until the drummer suddenly changes her rhythm. The drummer’s rhythmical change is not self-contained; it also changes the performance’s overall rhythmical properties. In particular, there will be greater rhythmical complexity than if all of the performers had continued to play the same motif. This complexity can then elicit a change in other players: do they maintain the complexity, or do they alter their rhythm to either enhance or diminish that complexity? In this way, one set of actions, properties, and organization belonging to the performers can change another subset of actions, properties, etc. only via the performance properties.
            It will be useful to economize our language at this point. We have three crucial elements in this structure:
(A)  Interventions: the actions, properties, and/or organization that change the performance properties;
(B)  Responses: The actions, properties, and/or organization that are changed via the performance properties; and
(C)  The performance properties.
Using these labels we can summarize our view. An event is a musical performance if and only if it possesses performance properties such that:
  1.  Interventions are possible. 
  2.  It is possible for interventions to affect responses only via performance properties.

    3.    Objections, with replies

Now, you may not be convinced. In particular, I will address two potential objections. First, doesn’t this way of thinking of musical performance rob it of its emotive and non-analytical qualities (§3.1)? Second, what happens with solo performances (§3.2)?

3.1.        Isn’t this soulless?

Even seeing the sentence “Musical performances are mechanistic systems” may offend the sensibilities of those who think of music in more emotive, non-analytical terms. I offer four replies. First, the terminology of systems and mechanisms is something that I inherit from philosophers of science, e.g. Carl Craver and Peter Machamer, but very little hangs on this choice of words. I welcome alternative terminology that is less offensive to the touchy-feelies of the world J.  For instance, since biologists frequently regard organisms as mechanistic systems composed of parts (organs, cells, and molecules), I could just as easily describe my view as thinking of music as an “organic whole,” which perhaps sounds less incongruous to those who dislike talk of mechanistic systems. Second, recall that the ultimate goal is to clarify the concept of a successful free music performance. Whatever the value of thinking emotively and non-analytically, it is frequently an ineffective tool when clarity is our aim. Third, it should be recognized that the fact that a system can be explained mechanistically is entirely compatible with taking emotions and non-discursive modes of thought as real phenomena. Indeed, the psychology of emotion rests on exactly this assumption. Finally, let’s first see how the whole approach works, and then decide if it’s inadequate to the task of clarifying a successful free music performance.

3.2.        Solo performances

I have argued that performances are mechanistic systems only if it’s possible for one subset of performers’ actions, properties, and organization (interventions) to affect another subset of performers’ actions, properties, and organization (responses) only via performance-level properties. At first blush, this would appear to require every performance to involve at least two performers. “However,” my critic continues, “that is surely incorrect, as there are solo performances.” Hence, a critic has a seemingly plausible argument that solo performances are not mechanistic systems.
            However, this rests on a confusion of what interventions and responses are. They are, first and foremost, sets of actions, properties, and (modes of) organization. They are not sets of performers. So multiple actions, properties, or modes of organization are needed for a musical performance; multiple performers are not needed.
Once this is appreciated, it is clear that solo performances satisfy our two conditions for a musical performance. Specifically, an event is a solo musical performance if and only if the event involves only a single performer and a set of performance properties such that:
(1)   Interventions are possible.
(2)   Interventions that affect responses only via performance properties are possible.
As before, condition (1) remains uncontroversial, as a soloist who played very different things would create very different performance properties. Given what we’ve said, (2) should seem unremarkable to anyone who has performed solo. For instance, suppose that our drummer from before is now a soloist who changes his rhythm mid-performance. Then clearly he has intervened on the musical properties. There is clearly a change in the performance’s properties, for its form now includes a rhythm change. However, the drummer can now respond: after the rhythm change, the drummer can decide to continue with the new rhythm, return to the previous rhythm, etc. and presumably this can be in response to the performance’s formal structure. Hence nothing about solo performances undermines my account of musical performances.

4.    The payoff thus far

So, why go through all of this? Recall the larger objective. We begin with a bunch of non-free musicians and are trying to get them produce a successful free music performance. Thus far, we’ve bracketed success and freedom, and simply tried to articulate what a musical performance is. Our musicians have thus come to see that a musical performance is a system with a particular set of properties, namely performance properties that can the musicians can intervene upon and to which they can respond.
            Now, let me suggest where I want to go next: different genres or kinds of music can be individuated by the norms that govern their interventions, responses, and performance properties. Hence, we can start to get a sense of free music in virtue of its distinctive configuration of these three elements.

Appendix: Some puzzles about responses

Above, I noted that the following is a tricky claim to establish, but didn’t want to belabor people with the details:
Had the musical performance exhibited different properties, then the performers’ properties, actions, and organization would have been different.
To see the difficulty with this case, we’ll need to say more about what we mean by manipulability in MMC. Here is the way that Jim Woodward (roughly) characterizes the idea:
An ideal intervention I on X with respect to Y is a change in the value of X that changes Y, if at all, only via the change in X.
In our particular case, X is the performance properties, Y is the performers’ properties, actions, and organization. So we’re looking for some ideal intervention I such that:
I is a change in the performance’s properties that changes the performers’ properties, actions, and organization, if at all, only via the change in the performance’s properties.
Now, the tricky bit is this. If this is going to be practical advice for performers, then the ideal intervention, I, should be something that they can do. However, that would mean that I is a performer’s action. But that would mean that performances change performers’ actions only via changing performers’ actions, which makes the performance’s properties totally irrelevant.
            The solution to this puzzle emerges from the fact that performances are composed of more than one action, property, and/or mode of organization. Consequently, one action, property, or mode of organization can serve as an ideal intervention (I) on a performance property (X), while another such action, property, or mode of organization (Y) can be a response to that performance property. That is the view I presented above.

Tuesday, April 17, 2012

Understanding and (the neglect of) Fictionalism


So I’m going to go a bit off of the script here, and focus on the paper that I see emerging from these readings. This means talking a bit less about fictionalism, though the class will provide many opportunities to do so.

1.   Elgin vs. Mizrahi

Last week, we saw Elgin’s argument that understanding entails neither truth nor belief, but merely “true enough” acceptance. To repeat, her master argument was as follows:

(1)  There are some phenomena p such that for some propositions x:
a.          x is acknowledged to be untrue, and
b.         x figures in scientific understanding of p.
(2)  For all propositions x, if x figures in scientific understanding of some phenomenon, then x is epistemically acceptable (120).
(3)  There are some propositions x such that x is acknowledged to be untrue and x is epistemically acceptable.

Elgin provided several examples from scientific practice to support Premise (1). The one of greatest interest for this week is the following:

(4)  There are some phenomena p, explained by some idealized law L, such that the proposition L:
a.     Is acknowledged to be untrue, and
b.     Figures in scientific understanding of p.
(5)  Thus, (1) is true.

Example: Ideal gas laws posit molecular structure and interactions that never obtains.

As we had already anticipated, there is a reply to this, which Mizrahi (2011) nicely discusses. We never need to acknowledge that the ideal gas law is untrue, for we are only ever committed to “Gas g behaves approximately like an ideal gas under conditions C” (where C specifies conditions of temperature, pressure, and the like.) Since it is true (not merely true enough) that the relevant gases behave approximately like ideal gases, this proposition (rather than the ideal gas law taken straight) need not (indeed, should not) be acknowledged as untrue. Furthermore, once we are dealing with propositions that are acknowledged to be true, the appropriate attitude to adopt is one of belief. Hence, true beliefs are still on firm ground, and understanding is still in the running as a species of knowledge.

2.   Potential theses for a paper…

Mizrahi’s paper is bad news for us—we’ve been scooped! What do we do? Here are some interesting observations from last week’s conversation that make things appear less bleak:
(1)  Mizrahi only does this with the ideal gas law, we can do the yeoman’s work of covering all of the other kinds of felicitous falsehoods that Elgin catalogues (e.g. stylized facts, curve-smoothing, etc.) By itself this is probably not a terribly exciting contribution to the literature, but not completely idle either.
(2)  We had other strategies available to us. For instance, critiquing premise (2) of Elgin’s Master Argument. This is certainly worth exploring, though there seemed to be some consensus that this would amount to little more than battling over definitions about “epistemic,” “cognitive,” and “pragmatic.”
(3)  We briefly mentioned another strategy—that of actually granting that felicitous falsehoods could figure in understanding, whilst denying that this precluded understanding from being a species of knowledge. I see two ways of arguing for this strategy.
a.     Semantic strategy. One option is to defend the idea that: If q explains p is true, then q need not be approximately true.
1.     What would be exciting about this strategy is that it would actually challenge both Elgin and Mizrahi.
a.     Pace Elgin, one could have the true belief that q explains p, but this could be a felicitous falsehood.
b.     Pace Mizrahi, this would say that understanding need not even be quasi-factive, for q could be radically false and still provide understanding via its explanatory role.
2.     Turning back to knowledge, the key bit would be the following inference would have defensible premises:
a.     If S understands why p, then there is some q such that S knows that q explains p.
b.     If S knows that q explains p, then q explains p is true.
c.     If q explains p is true, then q need not be approximately true.
d.     So if S understands why p, then there is some q such that S knows that q explains p, and q need not be approximately true.
3.     The key is defending premise 2.c.
a.     One option might be the following:
                                                                                                     i.     Our concept of explanation already tolerates “true enough” explanantia.
                                                                                                   ii.     True enough propositions that are not explanantia (or at least do not play an explanatory role of some sort) do not figure in our understanding.
                                                                                                  iii.     Hence explanatory propositions (q explains p) are true (full stop), even if their components (q and p) can be true enough.
b.     Epistemological Strategy:  Another option is to argue that it is possible that S knows that q explains p, even if S only accepts that q explains p.
                                               i.     There is a precursor to this, namely Cohen’s An essay on belief and acceptance
1.     I haven’t read this yet.
                                             ii.     This argument would run as follows:
1.     If S understands why p, then there is some q such that S knows that q explains p.
2.     If S knows that q explains p, then S accepts that q explains p.
3.     If S accepts that q explains p, then S need not believe that q explains p.
4.     So if S understands why p, then there is some q such that S knows that q explains p, and S need not believe that q explains p.
Three questions arise: What exactly is the difference between the Semantic and the Epistemological Strategy? Second, which we do adopt? And in either case, how do we defend the respective core claim?

3.   Semantic vs. Epistemological Strategy       

Looking at the relevant foils to these strategies best illustrates the difference between them. One foil is what we might call the “vanilla version” of an account of explanatory knowledge. That view would hold:
(1)  If S knows that q explains p, then q explains p is true.
(2)  If q explains p is true, then q and p are true.
(3)  If S knows that q explains p, then S believes that q explains p.
Elgin’s reasoning seems to be as follows. She grants the preceding three assumptions, then adds:
(4)  One can understand without having a true belief that q explains p.
From this, she infers:
(5)  So understanding is not a species of knowledge.
Let us assume, for simplicity’s sake, that if one understands then one has some sort of attitude about an explanatory proposition q explains p. Now, our endgame is to deny (5), but granting that this argument is valid, this means that we must deny one of the premises (1)-(4). The Epistemological Strategy denies (1) and (3), but is squarely committed to (4). By contrast, the Semantic Strategy denies (2), but can remain agnostic, if not tilted towards opposition, about (4). That’s the difference.

4.   Which Strategy?

            But which strategy should we pursue? There is a temptation to pursue both, but I think this is a mistake that many young researchers make. Since either would be interesting on its own, it would be far better to develop one really solidly, and then save the other for your next project. This is partly strategic: that’s two lines on your vita rather than one! But it’s also good for generating the best paper possible: given finite resources, it’s better to have one well developed position rather than two worse developed positions.
            OK, so which one? Well, part of this is just wherever your gut takes you—where do you see the stronger argument? With which literature are you more comfortable? But that’s not the whole story. There is also a question about the logical interrelations between (1)-(4). Specifically, Elgin’s critique of the vanilla version shares many assumptions with its target—namely (1)-(3)—but Elgin clearly privileges (2):
(6)  There are understanding-providing explanations of the form q explains p where either q or p is false.
(7)  By (2) and (6), q explains p is false.
(8)  By  (1) and (7), S does not know that q explains p.

(9)  If q explains p is false, then the appropriate attitude to it when it provides understanding is not belief.
(10)        By (7) and (9), the appropriate attitude to understanding-providing explanations is not belief.
(11)        So by (3) and (10), S does not know that q explains p.

So, (2) and (6) are the most basic premises in Elgin’s argument. Note that rebutting (6) is just to retrace Mizrahi’s steps, so rebutting (2) is really the way to shed new light on this conversation. This means that we should pursue the Semantic Strategy.

5.   How to defend the Semantic Strategy?

            What’s cool about this is that we can answer the last of our three questions. We can use Elgin’s same arguments to defend our challenge to (2)—but draw a very different and indeed more modest conclusion. On this view, we agree with her that false propositions figure in explanations (and other chunks of discourse/bodies of information), but this does not preclude understanding from being a species of knowledge. Thus, we can know that q explains p even if we don’t know that q (because, e.g. q is false).
            Next, we need to motivate this position. Here is at least one suggestive strategy: what if antirealism is correct? Then understanding is surely not factive. Hence if Elgin is right, it would seem that antirealism precludes understanding from being a species of knowledge (Elgin actually makes this argument elsewhere). Moreover, everything Mizrahi says indicates that he agrees with Elgin on this point. So, if it is possible to be an antirealist and maintain that understanding is a species of knowledge, then both Elgin and Mizrahi are incorrect. It is possible to be an antirealist and maintain that understanding is a species of knowledge (namely in those possible worlds where the Semantic Strategy succeeds). So both Elgin and Mizrahi are incorrect.
            So, there’s an outline of a paper somewhere in this. It goes something like this:

What is the thesis? Because true explanations can have false explanantia, understanding can be a species of knowledge.

What is the argument?
I.               If Elgin’s argument is sound, then understanding is not a species of knowledge.
II.             Elgin’s argument is not sound.
a.     Defend the idea that q explains p does not entail q.
III.           If Mizrahi’s argument is sound, then understanding is a species of knowledge.
IV.           Mizrahi’s argument is not sound.
a.     Defend the idea that antirealists are not opposed to understanding.
V.             Present a sound argument that understanding is a species of knowledge.
a.     If S understands why p, then there is some q such that S knows that q explains p.
b.     If S knows that q explains p, then q explains p is true.
c.     If q explains p is true, then q need not be approximately true.
d.     So if S understands why p, then there is some q such that S knows that q explains p, and q need not be approximately true.