Wednesday, May 03, 2006

The Self-Refutation Argument Against Eliminativism

Having gotten into some exchanges on eliminative materialism, it seems time now to take a look at how the argument that eliminative materialism is self-refuting should go, but perhaps also how it should not go.

To review, eliminative materialism is best understood in terms of a typology, developed by the Churchlands themselves, of scientific reductions. A conservative reduction keeps the reduced item intact, but simply gives it an scientifically accurate description. By describing heat in a gas as its mean kinetic energy we are not eliminating the idea of heat, we are just giving it an accurate scientific description. A reforming reduction asks us to accept an altered idea of what the reduced object is, but it does not make sense to say that the relevant object does not exist.

However, eliminativists promise that successor concepts will emerge from a matured neuroscience that will replace the concept of belief, so that an argument against eliminativism that ignores the promise of successor concepts can fairly be accused of begging the question.

Consider for example this argument.

1. The eliminativist sincerely utters, "There are no beliefs."
2. So, the eliminativist believes that there are no beliefs.
3. So eliminativism about beliefs involves realism about beliefs.
4. So eliminativism is incoherent.

This argument ignores the eliminativist claim that belief-successors will emerge. Eliminativists are never clear about whether "sincere utterance" will be retained in the brave new eliminativist world, or whether it will itself be eliminated and replaced with a successor.

Lynne Rudder Baker was, it think, the first philosopher to develop the argument against eliminativism in a way that takes into consideration the promise of successor concepts. How are these successor supposed to work. She offers three criteria for what they must accomplish:

i) Without appeal to the content of mental states, the alternative account of assertion must distinguish assertion from other audible emissions.
ii) The alternative account of assertion, against without appeal to the content of mental states, must distinguish sounds that count as assertiojn that p rather than assertion that q.
iii) The alternative account of assertion must at lest have conceptual room for a distinction between sincere assertion and lying.

BDK, our resident EM defender, answered the last of these questions as follows:

To lie is to know X is false, but to assert X anyway. The EM advocate would just say that knowledge is a property of internal nonpropositional representational states that can be true or false, or if you prefer, can provide a better or worse fit to the world. This, of course, is the positive story Churchland has been developing with his state space semantics, or recently he's been calling it 'domain portrayal' semantics.



or again

Chapter 2 of Paul's new book covers, in detail, his theory of conceptual content and concept acquisition in neural nets.

He also spends about five pages explicitly discussing the realism question. He lays out in more detail how he thinks we can get more or less accurate internal maps of the world which we use to navigate (and we do know that brains use internal maps to steer about in the world: and the maps aren't just of space but more abstract features of the world). He also admits that there is no way to stand outside our own conceptual framework to compare it to the things themselves and see how good the fit is between the two. He discusses what this means for the pragmatic realist like himself.

I'm not sure when it will be out, but the working title is 'Outer Spaces and Inner Spaces: The New Epistemology.'

It is pretty clear that his theory does not employ propositional contents. Even Fodor agrees with this. If you wanted, therefore, to say that truth and falsity are properties of Paul's conceptual spaces, then you would have to say that truth and falsity can be a property of nonpropositional contents. Otherwise, some other normative target is required for the conceptual spaces, and I think Paul has found a natural, and reasonable, one.


But here I start to have trouble. The trouble I have is that these states pick out propositions. They can be true or false in virtue of their relationship to their propositional contents. The fact that they don't satisfy Jerry Fodor's idea of what a proposititonal attitude is supposed to look like does not strike me as especially critical.

Either these states pick out propositions or they do not pick out propositions. If they pick out propositions, then they have a property (picking out p) which is going to be inconvenient to the project of reducing everything to physics and unifying science. Whether they look like sentences in the brain or not makes no never mind to me, if the state of a persons picks out a proposition well enough so that we can distinguish between assertion and nonassertion, then we have a belief. If on the other hand, no states of the person pick out propositions, then there is no way that there can possibly be successors that do the work that Baker, quite rightly, has pointed out that they must do.

http://www-unix.oit.umass.edu/~lrb/work/publications.html

2 comments:

Blue Devil Knight said...

Your editing is great: you have made me sound more coherent than I actually am.

The key is this claim in your response:
The trouble I have is that these states pick out propositions. They can be true or false in virtue of their relationship to their propositional contents.

It is not obvious to me that the neural states pick out, or have, propositional contents. (I'm not sure what it means to 'pick out' a proposition).

Does a map of the streets of San Diego have propositional contents? I am frankly not sure. If you think they do, then you'll think that Churchland's theory is just an implementation of propositional attitudes. If they do not, then you'd be an eliminativist. Note this isn't the same as saying that we can make statements (in language) about the map that are true or false. We can do the same with our phenomenal experience, which you have admitted has nonpropositional contents. The key, then is, is the maps representational format propositional or nonpropositional?

Churchland thinks they are nonpropositional, and that neural spaces have the same type of content as maps. To the extent that we can judge a map's accuracy, it is based on the relative locations of points on the map, not the properties of individual points.

You could say that the objective spatial coordinates on the map (e.g., this point X,Y on the map means you are at location X in the world) don't have this relational feature, but it is again not clear that this is the right way to speak of a continuous spatial map: couldn't we just as easily say that such a representation uses a continuous space, which is inimical to propositions, which quantify over elementary meaningful symbols?

One final thing to think about. Leeches have internal maps of the location where they were touched, and they use these neuronal maps to guide their behavior with respect to the stimuli. Would this imply that leech nervous systems have propositional contents because we can evaluate them wrt their accuracy? If so, then we should be applying propositional attitude psychology to leeches. If not, then why, in more complicated organisms, do maps gain propositional contents? This is not pie-in-the sky: I mapped out the leech representation of tactile space for my doctoral dissertation.

I realize it's easy to make fun of this stuff when we talk about leeches, but one of the assets of Churchland's view is that he starts out with the realization that language is a recent, unusual, and specialized aspect of neuronal function. Our neuronal theory of behavior should be more general, applicable to many creatures with nervous systems, but with some parameters tweaked to give rise to the cognitive abundance we see around us.

Amy Melser said...
This comment has been removed by the author.