Saturday, June 16, 2012

Stanley on Cognitive Neuroscience

I'm still working my way through the required reading for Jason Stanley's upcoming Meisterkurs at Berlin School of Mind and Brain.  I'm currently reading Chapter 7 of his Know How (2011).  At the moment, he's talking about how cognitive neuroscientists conceive of the distinction between declarative and procedural knowledge.  His argument is that the distinction has no implications for the question of whether knowing how is propositional in nature.

Stanley makes some very questionable moves here.  First, he discusses the distinction between declarative and procedural memory.  He is appealing to Gabrieli, 1998, p. 90:

nondeclarative or procedural kinds of memory encompass the acquisition, retention,and retrieval of knowledge expressed through experience-induced changes in performance. These kinds of memory are measured by indirect or implicit tests where no reference is made to that experience. Skill learning, repetition priming, and conditioning are classes of implicit tests that often reveal procedural memory processes dissociable from declarative memory.
The distinction is between memories which can be "consciously and intentionally recollected" (ibid., p. 89) and those which cannot, but which affect intentional behavior in other ways.  Stanley objects that we can take this account and still think of procedural knowledge as propositional.  He writes (p. 211):
Procedural knowledge is knowledge that is typically expressed through an increase in
skill. But knowledge of propositions could easily be expressed through an increase in
skill. My belief that I should catch a fly ball by positioning my body in a certain way
could become knowledge by practicing catching fly balls in that manner. The practice of
catching fly balls in that manner would give me proprioceptive evidence that my belief is
true. Unless one thought (absurdly) that proprioception is not a source of evidence, it is
obvious that propositional knowledge can be expressed through an increase in skill.
I want to give this argument careful attention, because it seems invalid.  More, it actually hurts Stanley's position.

One of Stanley's premises is that practice (e.g., the practice of catching fly balls in a particular way) can give one evidence in support of one's belief (e.g., that a particular method or rule is helpful when catching fly balls.)  I see nothing wrong with that premise.  So it does seem that practice can help turn belief into knowledge.  So, clearly, proprioception can be a source of evidence. But Stanley also has this premise:  that, given that proprioception is a source of evidence, it is obvious that propositional knowledge can be expressed through an increase in skill.

Not only is it not obvious that propositional knowledge can be expressed through an increase in skill; this premise is not consistent with the first premise, which states that propositional knowledge can be gained by observing that same increase in skill.  We can gain propositional knowledge by observing how we exercise a skill, as Stanley says, but that cannot mean that the exercising of the skill expresses the propositional knowledge it was supposed to justify.  If one has expressed their propositional knowledge that X by showing an improvement in their skill, then one would certainly not need to (and would not even be able to) use proprioception of that skill to move from mere belief that X to knowledge that X, because one would already have knowledge that X.  The fact that we can gain knowledge that X from observing our skilled performance implies that the skilled performance cannot be a manifestation of our knowledge that X.  Thus, either it is not a manifestation of knowledge at all, or it is knowledge of some other variety.

After this curious misstep, Stanley attempts to argue that, just as the two kinds of declarative knowledge (semantic and episodic) can be analyzed as different kinds of propositional knowledge, so we might expect procedural (nondeclarative) knowledge to be just another kind of propositional knowledge.  He suggests that they might be propositions which describe ways of doing things.

Now, obviously, we can have semantic and episodic memories about ways of doing things.  So why oppose that category of knowledge to declarative knowledge?  He mentions something about amnesiacs:

the procedural knowledge that an amnesiac can acquire is garden-variety
knowledge of a special class of propositions, those describing activities. For example, an
amnesiac who can acquire a pattern-analyzing skill can acquire knowledge of
propositions involving ways of doing that skill, but cannot acquire knowledge of many
other kinds of propositions, such as propositions about her past experiences. Even if
differences in how knowledge is implemented do correlate with kinds of content, the
contents of procedural knowledge are still propositions.

However, this is not an argument for the conclusion that procedural knowledge is propositional knowledge.  It is merely just stating that conclusion, and in a very counter-intuitive way.  For, again, why can't propositions about ways of doing something be semantically or episodically recalled?  And why should we think that what she has acquired is propositional knowledge?

Stanley also suggests that the way the declarative/procedural distinction is framed in AI might be applied here, so that we can suppose that what cognitive scientists call "procedural knowledge" might involve the sorts of representations that AI researchers call "declarative knowledge." Since Stanley thinks the AI distinction is not between two different types of knowledge, but between two different ways of implementing propositional knowledge, Stanley concludes that the cognitive neuroscientific distinction between declarative and procedural knowledge is not necessarily a distinction between propositional and non-propositional knowledge.  Yet, as I argued in my last post, Stanley has gotten the AI distinction wrong.

Still, perhaps Stanley could have a valid argument here.  The fact that cognitive neuroscientists distinguish between declarative and procedural knowledge on one level (the level of conscious access) does not exclude the possibility that these different kinds of knowledge might both have underlying propositional structures.  The AI distinction can be of help to Stanley here.  He can suppose that we do have declarative knowledge (in the AI sense) underlying what cognitive scientists call procedural knowledge.  It's just that we lack conscious access to the propositional contents.  In that case, however, the relation between a person and a proposition which, on Stanley's view, constitutes propositional knowledge must consist at a sub-personal level, through the base computational structure of the brain.  It is a relation between a person and that person's programming, and not a relation at the personal level of self, agency and intellect.  This is assuming, of course, that people are at base computational machines which implement programs the way AI declarativists think, and that is far from given.

At his most modest, Stanley only claims that we can interpret the cognitive neuroscience literature as being consistent with a view of procedural knowledge as a variety of propositional knowledge.  However, he is not always at his most modest.  For example, his conclusion on page 213 is a little shocking:
The obvious way to take the neuroscience literature is in the terms I have described. The
content of procedural knowledge is propositional, but involves different kinds of
propositions than stock cases of declarative literature.
Why is that obvious?  The cognitive neuroscience literature is continuous with the psychology literature, which for decades has drawn the declarative/procedural distinction in a very different way.  For example, Anderson (1976, p. 78) clearly identifies procedural knowledge with "knowledge about how to do something" and declarative knowledge with "knowledge of facts about the world."  Clearly the most obvious way of reading this distinction is not to take it as between different varieties of propositional knowledge, unless one were to think it obvious that all knowledge was necessarily propositional in some shape or form.

Stanley says (p. 213), "there is nothing in cognitive neuroscience that entails that procedural knowledge is not a species of genuine propositional knowledge."  I am not sure what Stanley means by "genuine propositional knowledge," nor am I convinced that this statement is a faithful representation of the field of cognitive neuroscience.  Perhaps Stanley can find a way to interpret the cognitive science literature as allowing for a propositional reading of procedural knowledge.  Perhaps there is enough support in the field for a declarativist view of knowledge implementation.  However, he has not shown that the literature is entirely amenable to such a view.  He has, on the contrary, engaged with an incredibly small portion of the literature.

Later in Chapter 7, Stanley discusses verbal reports of knowledge how.  He mentions a boxer who cannot verbalize his way of fighting a southpaw very well, but can vaguely identify it by performing and saying, "This is the way I fight against a southpaw" (Stanley, p. 219).  Yet, Stanley claims that reports such as this are not sufficient for declarative knowledge.  Being able to verbally declare that one's behavior is a way of doing something is not, in Stanley's view, a declarative competence.  Why not?  Apparently, because an indexical is given, and not a detailed description of the behavior.  But so what?

Clearly the linguistic centers of the brain must be involved.  There must be some linguistic competence there and it must be connected to whatever procedural knowledge is being implemented.  Yet, the linguistic competence is not identical to the boxing skill.  So clearly, there is declarative competence which can be distinguished from the procedural competence.  And clearly, being able to verbally identify a way of fighting against a southpaw is quite different from being able to implement that way of fighting against a southpaw.

It's odd that Stanley claims the boxer is not exhibiting declarative knowledge, when on the previous page, he observes that there is no neat way of characterizing declarative knowledge as cognitive scientists conceive of it:

Just as it is an open question how to characterize the notions of “implicit” and
“explicit” knowledge, it is equally an open question how to characterize the somewhat
technical notion of declarative knowledge as it occurs in cognitive neuroscience.

If the nature of declarative knowledge is still an open question in cognitive neuroscience, then why prejudicially assume that verbally identifying one's skills is not an exercise in declarative knowledge?  I don't see any argument for this in Stanley's tome.  He supposes that declarative knowledge might be knowledge that can be accurately and informatively described.  But isn't the boxer informing us when he identifies his skill verbally?  Isn't it an accurate statement?  Stanley (pp. 220-221) claims that what counts as informative and accurate is context-sensitive, but he has no doubt that the boxer's statement simply does not count, full stop.  Again, I don't see an argument supporting this counter-intuitive assertion.

I'm still working through Chapter 7 of Know How, and then I have Chapter 8 waiting, as well.  So, more to come.  But first, a little reflection on methodological assumptions.

On the one hand, there is the question about what various attributions of knowledge entail.  What do people mean when they attribute knowledge how to do something?  On the other hand, there are questions about the kinds of knowledge we actually have, and if all of them are propositional in one way or another.  Do they involve having conscious access to propositions?  Do they involve having organizational features which unconsciously utilize propositional knowledge states?  Do they imply an ability (or counterfactual ability) to verbalize or linguistically identify facts?  We might find that all kinds of knowledge involve some kind of relation between a person and a proposition, but that different kinds involve very different sorts of relations.  Yet, this will not mean that attributions of knowledge how to do something are actually attributions of knowledge that something is the case.  Not if we find that people attribute these different kinds of knowledge in independent ways and if they imply different relations between persons and propositions.

The fact that all human knowledge might be propositional in some base, computational way will not mean that knowing how is a species of knowing that.  And a linguistic analysis of the features of knowledge attributions is not the sort of study which should help us figure out how our brains actually work.  There is no reason I can imagine to suppose that the linguistic features of knowledge attributions would reflect the underlying computational scheme of our cognitive processing, or that we can infer from the grammar of the constructions to the operational structure of our brains.  It is much more likely that the meaning of these terms is related intimately to how they are used, and that they are used to distinguish between behaviorally distinct capacities.  This is what the folk distinction between knowing how and knowing that is about.  It follows something more or less like the psychologist's distinction between procedural and declarative knowledge, and not something like the AI distinction between different ways of implementing knowledge.