Tuesday, September 4, 2012
Brain hacking and mind reading
Over the last week or so several news stories have appeared (e.g. here and here) suggesting that it is technologically possible to “hack” the brain and extract from it PIN numbers, credit card data, and the like. This naturally raises the question whether such a possibility vindicates materialism. The short answer is that it does not. I’ve commented on claims of this sort before (here and here) but it is worth revisiting the issue in light of what I’ve said in recent posts about how the Aristotelian-Thomistic (A-T) philosopher understands the relationship between thought and brain activity.
Recall the analogy I drew in my recent BioLogos Forum piece between sentences and thoughts. Any token of an English sentence like “The cat is on the mat” has physical properties apart from which the token could not exist -- such as, in the case of a written or printed sentence, the letters and their shapes and sizes, the ink in which they are embodied and its various chemical properties, and so on. To damage or destroy these physical features is to damage or destroy the sentence token. Yet the meaning of the sentence, which is obviously no less essential to it than its physical properties are, is in no way reducible to, supervenient upon, or in any other way explicable in terms of those physical properties. The physical and the semantic, material and immaterial, form a seamless unity.
Similarly, for the A-T philosopher, human thought in normal circumstances is a seamless unity of the material and immaterial, the physiological and the psychological. As Ric Machuga puts it in In Defense of the Soul, “souls are in bodies the way meaning is in words” (though I would emphasize that this is only an analogy, and I should also caution the reader that Machuga’s presentation of A-T hylemorphism is problematic). I’ve noted recently how the content of a thought cannot be identified with mental imagery of any sort, nor entirely explained in neurophysiological terms. Yet our intellectual activity typically takes place via imaginative and material media. Even when our thoughts are at their most abstract, they tend to incorporate imagery of some sort -- hence (say) we think of visual or auditory images of numbers or shapes when we work through mathematical problems, or of the words for abstract concepts when we think philosophically, even if the thoughts cannot be reduced to these exercises in imagination. (This is why it is easy to fall into the error of identifying thinking with the having of mental imagery.) As Aquinas writes in the Summa Theologiae:
Although the intellect abstracts from the phantasms, it does not understand actually without turning to the phantasms. (I.85.5)
[I]t is clear that for the intellect to understand actually, not only when it acquires fresh knowledge, but also when it applies knowledge already acquired, there is need for the act of the imagination and of the other powers. For when the act of the imagination is hindered by a lesion of the corporeal organ, for instance in a case of frenzy; or when the act of the memory is hindered, as in the case of lethargy, we see that a man is hindered from actually understanding things of which he had a previous knowledge. (I.84.7)
[I]n the present state of life whatever we understand, we know by comparison to natural sensible things. Consequently it is not possible for our intellect to form a perfect judgment, while the senses are suspended, through which sensible things are known to us. (I.84.8)
From the A-T point of view, then, it is hardly surprising that neuroscience has uncovered intimate correlations between neural activity and mental activity, or that damage to the brain can severely impair thought -- any more than it is surprising that if we physically damage a sentence, its ability to convey its propositional content is diminished or destroyed despite that content’s being irreducible to the sentence’s physical properties. For the A-T philosopher to acknowledge that there is a physiological component to thought is not to make a desperate concession to modern scientific advances. On the contrary, it is merely to reaffirm something that Aristotle and Aquinas themselves already recognized.
What the A-T view should lead us to expect, then, is precisely that we should be able to “read off” the mental from the physical to some extent, though only to some extent. And ironically, this is precisely what materialist writers like W. V. Quine, Daniel Dennett, Bernard Williams, and Donald Davidson say insofar as they affirm that the meaning of our linguistic utterances and thoughts is not fixed or determined by any set of physical facts. The difference is that since they are materialists, they conclude that there just is no “fact of the matter” about what our utterances and thoughts mean; whereas the A-T philosopher, who holds that the claim that there is no “fact of the matter” about meaning cannot be coherently made out, concludes that meaning is something immaterial. Both sides would agree that neither the methods described in the articles about “brain hacking” linked to above nor more sophisticated methods are ever even in principle going to get you to a strict predictability of the content of thought from the physical facts, even if they get you arbitrarily close.
The reason, it must be emphasized, is not epistemic but metaphysical. It has nothing to do with how many of the physical facts we can know but with what those facts would by themselves entail, even if we knew them down to the last detail. Nor does the point have anything essentially to do with complexity, precision of measurement, or with whether the relevant physical facts are inside the nervous system or in a thinker’s physical environment. Even given the simplest and clearest physical symbol possible and the most detailed, exhaustive description possible of that symbol’s causal and other physical relations to the entire material universe, we would still not know with certainty what that symbol means because the entirety of those physical facts by themselves simply would not specify a unique meaning. It’s not that the physical facts would entail that such-and-such is what is meant, but we couldn’t know enough about these facts to find out for sure that it is such-and-such that is meant; it’s that they would not entail this in the first place. (Again, the point has nothing essentially to do with whether or not one is a materialist -- Quine and Co. would say the same thing. The dispute between these materialists and their critics is over whether there is something additional to the physical facts that does determine meaning.)
The analytical Thomist philosopher John Haldane provides a useful illustration:
Every triangle is a trilateral and vice versa, and in some manner possession of the one property necessitates possession of the other. Yet triangularity and trilaterality are not the same attribute, and it takes geometrical reasoning to show that these properties are necessarily co-instantiated… To the extent that he can even concede that there are distinct properties the naturalist will want to insist that the causal powers… of trilaterals and triangulars are identical. Thus he cannot explain the difference between the concepts by invoking causal differences between the members of their extensions (as one might seem to be able to account for the difference between the concepts square and circle). (J. J. C. Smart and J. J. Haldane, Atheism and Theism, Second edition, pp. 106-7)
As Haldane notes, the problem is completely general:
For any naturally individuated object or property there are indefinitely many non-equivalent ways of thinking about it. That is to say, the structure of the conceptual order, which is expressed in judgments and actions, is richer and more abstract than that of the natural order, and the character of this difference makes it difficult to see how the materialist could explain the former as arising out of the latter. (p. 107)
In short, any set of material facts, including facts about the efficient causal relations between material elements, is indeterminate between the different determinate ways in which we might conceptualize them; hence the former cannot suffice to account for the latter.
So, suppose we really did have sentences encoded in the brain, and suppose even that those sentences were not the subtle kind posited by Fodor’s “Language of Thought” hypotheses, but were rather the crude sort we normally think of when we think of sentences. In particular, suppose that we found that the English words “The cat is on the mat” could be seen clearly written across the surface of a subject’s brain every time a cat entered the room and sat down on the mat. There would still be absolutely nothing in any set of physical circumstances that could possibly tell us for sure that what he was thinking was that the cat is on the mat. For one thing, the set of marks on his brain could in principle be a higher-order representation -- in particular, a representation not of cats on mats, but instead a representation of representations of cats on mats. Nor could what the guy says to us determine which of these the brain sentence in question really represents. For example, if he says “No, I’m thinking about cats on mats, not about representations of cats on mats,” we still need to know what that utterance meant, and whether it was itself really about cats on mats, representations of cats on mats, representations of representations of cats on mats, or whatever.
Note that the point is not that the guy might be lying to us, but that there is nothing about the physical properties of his utterance, or about his brain, or about his facial expressions or anything else, that could by themselves strictly entail that he means one thing rather than another even if he is not lying. Nor could the causal relations between his brain and the outside world tell us. For example, any regular causal correlation between the presence of a cat on a mat and the subject’s going into brain state B hardly shows all by itself that B represents cats on mats rather than representing representations of cats on mats, since the causal correlation in question could obviously be associated with either meaning. And any appeal to further causal considerations just kicks the problem back a level. (Again, there is nothing in this much that only a non-materialist could accept. It’s more or less just standard Quine-Dennett style indeterminacy stuff.)
As I have said, the problem arises at least in principle even in the simplest cases. But it is even more pronounced the further we get from thoughts about immediately present concrete objects and have to interpret thoughts of a highly abstract and theoretically loaded sort. Then the various possible “manuals of translation” -- to borrow the language of Quine’s famous “gavagai” argument -- become very unwieldy indeed. (We know this from everyday experience. It goes without saying that it is much harder to be sure one understands what another means when he is talking about some unfamiliar philosophical or scientific theory than when he is talking about the weather or the lunch you and he are sharing.)
So, while it may well be possible for neuroscience to give us increasingly effective technological means of guessing at what is going on in someone’s mind, this will always amount to something more like skimming a few dollars from someone’s bank account rather than emptying it. Between the brain and the mind, there’s enough metaphysical slack to prevent a complete hack.