Consciousness and the Overton window of science, Part IV

By Jonathan Birch

(See the other posts in this series here!)

Part IV: Correlation and Explanation

I’m really grateful to Hedda Hassel Mørch for writing such a thoughtful response. To rewind a little: my essay “Consciousness and the Overton window of science” was a series of reflections on the state of consciousness science, not a single argument. But the gist of it was: there is deep metaphysical disagreement among rival groups of scientists, and we should not expect neuroscientific evidence to resolve that disagreement. I added that IIT is distinctive in offering a metaphysical picture with an idealist flavour, a picture outside the “Overton window” of cognitive-neuroscience-as-usual. 

This raises a zoomed-in question about how best to interpret the metaphysics of IIT, and a bigger-picture question about the relation between metaphysics and science. Mørch makes insightful comments on both questions. 

In earlier work, Mørch sought to unify IIT with “Russellian panpsychism“, a picture on which at least some basic mental properties inhere in the fundamental nature of matter. The argument was that IIT, though not fully compatible with Russellian panpsychism, could be made compatible with it through small amendments. Mørch’s article convinced me for a while that IIT was best interpreted in this way. I started to have doubts, though, when I saw how vehemently Tononi and colleagues rejected the idea of emergence. Shouldn’t a Russellian panpsychist be comfortable with the idea that macro-conscious subjects are in a sense emergent, in so far as they emerge from the combination of many micro-conscious subjects? I spoke to Tononi earlier this year, and he confirmed that IIT is no Russellian panpsychism, in his view. 

So, what is the background picture, I thought, if not Russellian panpsychism? I came to the view that IIT is more Leibnizian than Russellian. It’s a theory on which conscious subjects are the bedrock of reality. The things we call substrates of consciousness—neurons, brain regions, brains, and even fundamental particles—have no “intrinsic existence”, only “derived existence”, existence for conscious subjects. This is close to a world of Leibnizian “monads”. It’s just that, unlike monads, the conscious subjects of IIT have irreducible causal power and interact with each other. 

But what does this mean for consciousness science, and especially for attempts to test IIT against other theories? 

***** 

Mørch suggests that we can cleave apart the “metaphysical” and “empirical” sides of IIT. The empirical side just says that conscious experience correlates with brain structures that maximize phi. This correlational claim is metaphysically neutral and can be tested with normal neuroscientific methods.  

This is line with several recent attempts to distinguish “stronger” and “weaker” versions of IIT. Confusingly, a number of different proposals now coexist, with slightly different senses of “strong” and “weak”. Rarely has a theory received so many compassionate attempts to free it from its professed core commitments. I can’t get too deep into that here. To keep things simple, I will talk about “full IIT” where I mean the entire thing, complete with metaphysical commitments, and I will talk of “correlational IIT” to refer to the claim that the neural correlate of consciousness is whatever brain region is a local maximum of phi. 

What to make of this idea? I don’t want to sound too pessimistic about the future of consciousness science. I’m reasonably hopeful that the field may be able to settle the “prefrontal cortex vs. posterior hot zone” debate over the long run. At the very least, I expect evidence to shift the dial one way or the other. I also think the clash between cortical theories more generally and the midbrain-centric theories of Bjorn Merker and Jaak Panksepp may be long-term resolvable. 

Suppose, then, we get to a point at which all agree that the neural correlate of consciousness is located in posterior cortical areas. The trouble is that we will still be faced with many different hypotheses about what exactly makes these areas special. Phi will be on the table, but I’d expect Lamme’s recurrent processing theory to still be on the table as well, along with Block’s view that ties consciousness to a form of fragile short-term memory. So, we will have a family of surviving “local” theories: some structural, some functional, some on which details of implementation really matter, others that are more to do with high-level functions.  

I doubt empirical evidence of a correlational type will select decisively among the various local theories. As is commonplace in other areas of science, we will instead have to think about the overall explanatory merits of the different theories and try to make an inference to the best explanation. But now, correlational IIT can be seen in advance to have no chance of winning that contest, because it avoids saying anything explanatory at all, and limits itself to positing an unexplained correlation. 

Full IIT, by contrast, is at least in the business of offering explanations. Indeed, its defenders tend to say it is the only good explanation on the market, and so is justified by a principle of “inference to a good explanation”—absent serious competitors, comparative evaluations are not really needed. Yet this hints at a mismatch between what counts as an explanation within the IIT fold and what counts on the outside. For most of mainstream cognitive neuroscience, the kind of explanation offered by IIT is hard to recognize as an explanation at all. Reading passages like the one I quoted, they are likely to say: an “explanation” on which neurons don’t intrinsically exist, but exist only in a derived sense, as operational substrates for conscious subjects?! What sort of explanation is that? IIT may be offering explanations, but they are not the sort of functional-mechanistic-reductive explanations that neuroscience-as-usual recognizes.  

The wider lesson is that our metaphysical background picture shapes the criteria for what counts (for us) as a good explanation of consciousness. This makes theory choice using the method of inference to the best explanation—a method that works very well in normal science, where a shared metaphysical background exists—an inevitably question-begging exercise in the presence of deep metaphysical disagreement. Each side sees itself as providing real explanations, while regarding the other side as merely claiming to be explaining consciousness, misleadingly. 

***** 

Having set out to write a “concluding post”, I feel a certain amount of pressure to extract an overall message. And I know some will want the message to be: Down with IIT! Make space for cognitive-neuroscience-as-usual to generate cumulative progress! 

But I’m uneasy with this. It sounds too much like an attempt to create a new “reputation trap”, a no-go area for junior scientists and grad students. On the contrary, I think the phenomenon of the “reputation trap” expresses a vice of the scientific character, not a virtue. When we dismiss heterodox ideas as too radical, we risk foreclosing very-high-risk, very-high-gain opportunities. A scientist should be able to develop such ideas without being shunned by their field. When shunning happens for this reason, the incentive structure of science is going wrong, penalizing heterodox thought too severely. Aggressively policed Overton windows are a not a good thing. 

So, my preferred message is a gentler one: let’s accept that deep metaphysical disagreement leads to entrenched disputes that empirical evidence cannot settle. Let’s accept, too, that consciousness is a topic that will always generate such disagreements. Their existence is not something to sweep under the rug, but something to learn to live with, uncomfortable as that may be.