Thursday, December 27, 2007

Physical Theory of Subjective Mental States... or Not

Today I'll be reviewing "Towards a Physical Theory of Subjective Mental States" by Sean Lee. This marks my first entry on a physics journal article. ...You can stop cheering now. I've been looking for something I can post to use the BPR3 icon and, sadly, this isn't it. As you may have noticed from the above link, I found it on arXiv, so it doesn't satisfy the "peer-reviewed" criterion of the BPR3. And, even though I said it's a physics article, it doesn't really have anything to do with... well, let me not get ahead of myself.

The paper has been submitted to the physics-meets-philosophy journal Foundations of Physics. I don't know if this article will actually appear in the journal, as I am not sure what sorts of articles they take. I do have a new respect for the editor, Gerardus 't Hooft, based on this quote from the journal's homepage:

We receive numerous submissions from people who venture to attack the most basic premises of theories such as Special Relativity, but instead only succeed in displaying a lack of professional insight in how a physical theory is constructed. I suspect that some of these people may have been working somewhere in an attic, deprived from daylight for decades, determined only to reemerge with a Theory of Everything in their hands.

Anyway, "Towards a Physical Theory..." begins by introducing the hard problem of consciousness. In short, neuroscientists can, in principle, explain all the steps involved in the process of a stimulus to a person (say, a rock falling on one's foot) causing a response (for instance, jumping up and down, yelping) using nothing but physical interactions (neurons relay pain signals, we rub the area to stimulate non-pain receptors, we jump to shift weight to the other foot, etc.). I don't mean to say that they can explain every step now, but extrapolating into the future, it is reasonable to say they will be able to in the future. However, they cannot yet, and may never be able to, explain why we feel something (i.e. why the rock hurts one's foot). Currently we cannot explain first-person subjective experiences. Or, at least, that's what Sean Lee claims (not that I'm challenging that claim, I just haven't checked it out thoroughly). The activity of the brain is necessary and sufficient to cause mental phenomena, but we're not sure how "Neural Correlates of Consciousness," as they are called, produce subjective experiences.

In order to probe the barrier between the brain states and the mental states, Sean defines a mapping.
Within the context of a physical science, [the assumption that subjective states correlate to physical states] translates into requiring an unambiguous mapping M between whatever is identified as the space of physical states P with whatever is identified as the space of subjective states Q. We'll call this rather underappreciated assumption the mapping principle
M: Q→P
What is this mapping? What are the details? Lee doesn't seem to care much.
That is, the mere claim that subjectivity is at least correlated with physical states implies the existence of a well-defined mapping between the two.
In fact, Lee not only doesn't care what M is, he doesn't care about P either. P can be whatever you want, just some sort of physical description of the brain's state. With infinite precision.
States in P would [in a classical theory] be given by the distributions, momenta, and valences of all semi-classical atoms within and near the organism (by the generally presumed chain of supervenience, this entails all the understood rules of thermodynamics, chemistry, biochemistry, cell biology and the like). principle a purely quantum mechanical description of the physical state should be at least equally valid (if not manifestly illuminating). In this case P might be considered to be a space of state vectors |Ψ>, or perhaps alternatively, a relevant subset of measured observables {O1, O2,... ON}.
This sounds to me, if not an actual violation of the uncertainty principle, at least absurdly impractical. Knowing the state of every cell in the brain is completely implausible; knowing the position and momentum of each atom in any one cell is literally impossible. Trying to combine the two is simply laughable. It doesn't matter to Lee that these ps within space P can never be found, only that they are in principle mappable to the real numbers. Whatever. Moving right along.

The real problem for this paper (as if knowing complete information about the brain wasn't problem enough) is trying to find q, a subjective state within space Q of subjective states. You see, there's nothing we can measure to find q. After all, anything that can be measured is physical, thus those measurements provide information about p, the physical state. So how do we get information about q?
The inherently private nature of subjectivity forces us to accept that any operationally meaningful coding must be done by the experiencing subject. [emphasis original]
Wait, what?
Thus we come to the uncomfortable first conclusion that any meaningful operational definition that might rise to the standard of a science of subjectivity can only be given for a very limited number of subjects; namely for those that are willing and able to communicate with scientists!
No, you don't really mean—
In the case of subjectivity, however, the only relevant, operationally definable data that is accessible are from subjects possessing high-level language skills that are communicable to scientists.
Yes, Sean has done it. His method of "measuring" the mental state is to ask someone what their mental state is. He's pushed this completely out of the realm of any kind of scientific rigor and into anecdoteland. It gets better, though.

Not only does Lee remove his experiment from the realm of falsifiability, he sweeps away replication. He constructs a thought experiment in which a future scientist can induce a certain brain state in a subject, then performs the previously mentioned asking in order to collect "data" on the subjective state. However, once wouldn't be enough.
This pairing [of brain state and mental state] may be confirmed by repeatedly generating the state p0, each time immediately afterwards asking the subject:

'Are you certain that your experience now is any different than just a moment ago?' [emphasis original]
"Yeah, doc, it's different. Now I'm annoyed with you."

The procedure he outlines here is unable to collect real non-anecdotal data in the first place, but the attempt at replication and averaging introduces a new, constantly changing stimulus to the subject. As he asks the subject more and more times if his or her mental state is exactly the same as before, the experiment will be increasingly annoying and frustrating his subjects.

Non-objective (though I guess that was implied by the name "subjective state"), non-reproducible, non-falsifiable, this scheme has absolutely nothing going for it. The subject being studied is presumably supposed to say things like, "I feel hungry. This seat is a bit uncomfortable. I just farted, so I'm smelling that. You're standing too close and I'm getting creeped out." What about the innumerable other things the subject is thinking and can't put into words? What about the lag time between thinking and speaking? What about the fact that searching around for words to describe thoughts is a thought, and one that will probably not be mentioned without infinite regress?

Most importantly of all, what if your subject is lying, or even just keeping things from you? Perhaps he or she has just been cheated on by a spouse and doesn't want to talk about it, but that thought is constantly going through his or her mind. Perhaps he or she is uncertain about the future, but doesn't want to mention it to some jerk scientist. Maybe, just maybe, your subject doesn't care about your experimental protocol and will make up any old "mental state" just to screw around with your results. For all you know every test subject could be doing that. That's because this isn't science. There is absolutely no way to accurately obtain any data. You're relying on pure anecdote.

I hope I don't sound disrespectful; I don't mean to come off that way. I applaud any effort to investigate the nature of subjectivity empirically. However, I don't think the method described in this paper is worth our time. Truth be told, I don't think there is a method in this paper, just a collection of musings on what such a method might possibly look like. But I guess that's what I get for trawling arXiv for any paper that floats my way.


Ben said...

Just to be a little picky...

"(i.e. why the rock hurts)"

Don't you mean why the foot hurts? Rocks don't have feelings as far as I know.

Anywho, I happen to like arXiv for the most part. It has been great with astrophysics papers but I sometimes find that I have a hard time tracing the article to its original or considering journal. I also hope there is not too much of this kind of sloppy science floating around there. I knew it was a little more lax because they take papers that are submitted but not yet accepted for publication, but most of what I have come across has been scientifically sound.

Flavin said...

Foot thing: corrected.

And I only meant that a paper being on arXiv is no guarantee of quality. Hell, even being published in a journal isn't a guarantee of quality, so I don't think a web prepublishing service is full of 100% great, wonderful work. Not that everything is bad, either, but I rolled the dice by reading a random selection.

Blake Stacey said...

Oh, there's always more where that came from.