The Overflow Cup Runneth Over

There has been a lot of action on the overflow front lately! It started with papers by Ned Block and Dennett and Cohen in Trends in Cognitive Science (Block’s paper criticized, among others, my recent paper on this stuff). These articles spawned a response by me here and here, which I still stand by.

But now, having read the response from Kouider (which echoes his response given at his CUNY Cogsci talk) as well as the response from Overgaard and Block’s response to both of them in addition to Lamme’s response to Cohen and Dennett and their response in turn, it seems a couple points should be emphasized.

Accessed vs. Accessible
Block again and again says that his argument does not depend on inaccessible consciousness but rather on it being necessary that at any given moment there is some consciousness that is not accessed (but could be accessed at a different moment and so is not inaccessible tout court). There seem to me to be several issues worth considering here.

First is that there is a conceptual question about what it means to say that something is accessible but not accessed. One might think that you cannot know that something is accessible without it actually being accessed. Block and others respond that something is accessible when, roughly, it is globally broadcast. But then we might wonder why we ought to think that being globally broadcast is equivalent to being accessible. Aren’t their mental contents/states that are globally broadcast but which are not accessible? David Rosenthal has pressed this kind of question in conversation and I am not exactly sure what the neuropsychological answer is in this case. Is there any serious neuropsychological reason to think that global broadcasting is equivalent to being accessible?

Second one might worry about the ‘thin edge of the wedge’ implications of Block’s argument. If we can show that there is consciousness that is not accessed then it seems a short step to consciousness that is inaccessible in principle. And if it is true that there is a principled connection between the two then though it would be strictly speaking true that Block’s argument did not rely on inaccessible consciousness it would none the less still be appropriate to give a reductio of his argued for view in terms of absurdities in the view it leads to.

Finally, it does seem as though there is a principled connection between the two notions. Block argues that it is inappropriate to argue against the claim that there is inaccessible consciousness because he only requires that some consciousness not be accessed not inaccessible. But if one thinks just of a particular moment in consciousness leaving aside the next moment it is of course true that some consciousness is inaccessible. It is inaccessible at that moment. Block’s view is that at any given moment in your daily conscious experience there is, necessarily, some parts of your conscious experience that are inaccessible at that moment. Because of this his view really does have all of the problems associated with in principle inaccessible consciousness.

Given these considerations I don’t think that the appeal to the distinction between not accessed and inaccessible helps make Block’s case.

Kouider’s Data Count Against His Own View?
Block has said several times that Kouider’s own data counts against the no-overflow view. He says in his latest response,

According to the hypothesis Kouider et al. put forward, what is in consciousness before the cue are generic representations plus specific representations that are too sparse to provide the information necessary to explain partial report superiority. However, on their hypothesis one would expect a substantial error rate concerning the uncued items. However, Kouider et al. found the error rate to be small: their own evidence counts against them.

It really is not clear to me why Block thinks that one would expect a substantial error rate concerning the uncued items. He seems to be thinking that the no-overflow view is committed to only generic phenomenology before the cue but this is clearly not the case. It is compatible with the no overflow view that there is some specific phenomenology before the cue (just not all of the items as per overflow).

But even if one is not moved by this there is an obvious problem with the argument. The no overflow position maintains that there is enough information unconsciously processed to do the task. Subjects don’t make a lot of errors because that information was there whether consciously or not.

Falsifiability vs. Support by the Evidence
I think that Block is right that we do not want falsifiability as a=our standard here and that we need to evaluate theories holistically based on the widest swath of available evidence and theories available to us. Block thinks that there is some evidence that the kinds of unconscious processes necessary to sustain the no overflow view aren’t there. But this evidence is very weak and the jury is still out on this issue. In general the science is all over the place on this issue. There is partial evidence on both sides and no theory comes out on top on the basis of current scientific evidence alone. Hopefully this will change in the nearish future but at this point this is where it is.

Given this one might think that we should be agnostic about whether overflow is true or not but this doesn’t seem right to me. The overflow hypothesis is radical in that it postulates a kind of consciousness that cannot in principle be accessed (at that moment) and yet which is also for me in the way that normal accessed consciousness is for me. That is, I experience the unaccessed consciousness as mine without being aware that I do. How this could be so is deeply mysterious and perhaps in principle untestable with any known scientific methods. Barring prejudice in its favor we would need strong evidence indeed to accept such a notion.

Cognitive Access: The Only Game in Town

[cross-posted at Brains]

In Ned Block’s recent paper, published in Trends In Cognitive Science, he has defended his argument that perceptual consciousness overflows cognitive access from several recent objections (including mine). It is important that Block is defending overflow from cognitive access since he admits that perceptual consciousness does not overflow all access. Phenomenal consciousness consists in there being something that it is like for the subject of the experience and this suggests that there must be some kind of access to the experience. Block has elsewhere argued that some non-cognitive form of access can account for this but no account of non-cognitive access to date can explain what needs to be explained. Given this the anti-overflow position should remain the default until/unless we have much stronger evidence than what Block presents. Block suggests that there is a philosophical fallacy in the assumption that non-overflow is the default and in the insistence that we need strong evidence to overthrow the non-overflow position but this is not fallacious. It is the reasonable thing to do when you have very weak evidence that is consistent with two competing theories and one of those theories appeals to a mysterious place-holder concept while the other doesn’t.

Block suggests two possible forms of non-cognitive access. The first is a deflationary account and the second is a version of a self-representational theory.  On the deflationary account we are aware of our mental states just in the having of them, in much the same way that we smile our own smiles just by smiling. Recall that what we are trying to explain is how a particular experience comes to be for the person who has it. When I feel a pain, not only do I experience the painful quality but I also experience it as mine. How can the deflationary account handle this? The deflationary account applies equally well to any state that happens to be instantiated in the brain. We can say that we are aware, in this way, of a state in the LGN, for instance, but surely we don’t want to say that it is phenomenally conscious.

The same problems arise for a self-representational account. One kind of self-representational account, holds that the higher-order awareness is itself a part of the state that it represents. But this is a variant of a cognitive access theory. Block seems to want a notion of self-representation that amounts to the state in question merely being instantiated (in the way a color sample represents the color just by being that particular color). But then every state would be conscious since every state represents itself merely by being instantiated. In fact every representation self-represents itself in this way but we don’t want to say that sentences are phenomenally conscious!

These notions of non-cognitive access are too weak to distinguish conscious mental states from unconscious mental states, or from any kind of brain activity at all. On the other hand a higher-order cognitive representation explains how a mental state can be for me; I am representing myself as being in that state, in some suitable way, so I will naturally experience the state as mine.

Block endorses only the reasonableness of tentatively accepting the overflow conclusion. But until we have a notion of non-cognitive access that can explain how a mental state can be experienced as mine that is at least as satisfactory as that given by cognitive access we need much stronger evidence than what Block presents to accept overflow.