I have just returned from my favourite conference– the Association of the Scientific Study of Consciousness (ASSC), which this year held its 25th anniversary meeting in New York. This was an exciting trip for a number of reasons – my hotel was on Washington Square Park, where I spent my postdoc years at NYU; we held the second annual Perceptual Metacognition Satellite meeting the day before the main conference, and a number of my lab at UCL traveled out there to give talks and posters. I have a longstanding relationship with ASSC, going to my first meeting in Toronto in 2010, and then attending every meeting since except Buenos Aires in 2019, the year that my son was born. I was also on the Membership Committee for a number of years, won the Williams James prize in 2012, and was Executive Director from 2014-2020.
So I have a decent working knowledge of the Association, and know many of the regulars (I was also on the Scientific Committee for this year’s conference). But this year I felt something shifted. Which is why I am revisiting my blog to post a few reflections on the state of consciousness science.
First of all, I had a general sense that consciousness science was in excellent shape, with lots of potential. The co-directors of the NYC meeting, Dave Chalmers and Ned Block, pulled together a high-profile and varied program that was almost without exception of high quality, while not losing sight of the core focus on consciousness. The concurrent talks were universally excellent, reflecting the fact that this year there was fierce competition for abstract selection, and a record 700 attendees. This all led to a feeling (continuing on from last year’s also-excellent meeting in Amsterdam) that consciousness science is moving into a new, more mainstream era. As examples, the keynotes by May-Britt Moser, Doris Tsao and Yoshua Bengio came from outside the regular ASSC milieu, and were delivered by heavy-hitters in their fields (a Nobel prize winner, National Academy of Sciences member and Turing Award winner respectively). Strikingly none of them gave their regular keynote address. They each made a concerted effort to relate their mainstream neuroscience/AI research to foundational questions in consciousness. For instance, Tsao gave a beautiful talk on how basis function representations of face space relate to conscious experience as measured using binocular rivalry. These and other experiments hinted at important roles for feedback and feedforward sweeps in IT cortex initiating a switch between unconscious and conscious perceptual representations. Moser presented data obtained from baby rats that reveal that the grid cell system is already in place very early in life, prior to sensory experience – suggesting that a Kantian view in which we are born with spatial priors might be correct.
This was high-calibre mechanistic neuroscience of the sort that is commonplace at SFN or Cosyne. The difference here was that the bridge to subjective experience was also being taken seriously. There are considerable mutual benefits from this kind of cross-fertilisation between mechanistic/computational neuroscience and consciousness science. But it is fragile and hard-won – we should cultivate it carefully, and not take it for granted.
The students I talked to at the conference were similarly impressed, and saw this kind of work as the future of consciousness science. What is interesting here is that many of these experimental findings touched upon elements of theory that have been mainstays of previous iterations of ASSC (eg the “Great Debate” at last year’s meeting). Tsao’s findings appear partly consistent with the role of local recurrence and predictive processes in conscious experience. Bengio’s model of ineffability takes inspiration from the limited-capacity workspace of GWT. And so on. But the difference was that these theories were not being directly pitted against each other in a fight to the death. Instead they were being referenced as key components of a working model of the neurocomputational basis of conscious experience.
This contrasted with the write-up in the media of the public event on Thursday evening, when the Cogitate results were revealed. Cogitate is a large team-science project funded by the Templeton World Charity Foundation, and the first in a series of Accelerating Research on Consciousness initiatives. The idea is to adopt an “adversarial collaboration” model, where proponents of different theories team up to devise experiments that can tell apart different theories of consciousness. In this case, the project sought to test between global workspace theory (GWT) and information integration theory (IIT). I should say first up that I think the ARC projects are an important initiative, and their commitment to high-quality, well-powered open science will provide valuable data for the community for years to come. The energy, commitment and vision to lead such a collaboration should not be underestimated – the PIs on this project (Liad Mudrik, Lucia Melloni and Michael Pitts) have blazed a trail here and set very high standards for future projects in the field (myself and Axel Cleeremans are currently leading a similar initiative, following in the footsteps of Cogitate, to arbitrate between predictions of different higher-order theories).
From the presentations that evening and subsequent preprint, it became clear that the findings of this first experiment were inconclusive with respect to the theories (a paper on a second experiment remains in the works). This in itself is unsurprising – consciousness science is difficult, and a single experiment rarely is clear-cut. But as Lucia said on stage, we should applaud the bravery of everyone involved (from theorists to PIs to – especially – the trainees) in pursuing a team-science endeavour when the payoff is uncertain. I left the event feeling buoyed up and excited about his way of pursuing consciousness science.
I certainly did not leave the event thinking anything profound had been discovered about how consciousness works. This was all provisional, preliminary – another brick in the wall of data. But after I got back home I felt a growing sense of unease at how these findings were being portrayed in the media. There was an avalanche of high-profile coverage of the public event, fueled by it also being a vehicle for resolving a silly but entertaining “bet” between Dave Chalmers and Christof Koch about when the neural correlates of consciousness will be found. This coverage was not only in Nature, Science and Scientific American, but also in The New York Times and The Economist. It really hit home how much this reporting had cut through when my Dad started sending me links to articles he was seeing.
Their contents were alarming*. They suggested, some more strongly than others, that IIT had “won”. If I was a complete outsider to the field, and had idly picked up a copy of The Economist, I would have no reason to doubt that IIT is now a leading contender thanks to recent empirical findings. Imagine reading about a big-science experiment that is aiming to decisively test between two interpretations of quantum mechanics. Why should I question what is reported in Nature, Science, the NYT, or similarly respectable outlets? In her excellent Twitter thread after the meeting (and in an in-depth tutorial at the start of ASSC), Megan Peters pointed out the difficulty of accurately and responsibly communicating consciousness science to the public when it carries such widespread ethical and societal implications. Intentionally or not, the public event set off a chain reaction that led to an unfortunate mischaracterisation of the state of consciousness science, and one that may prove difficult to correct.
A core problem was that IIT was not really under test here – the extensive mathematical underpinnings of the theory did not enter into consideration at all. Instead, as the authors of the pre-registration acknowledged, the key hypothesis of a link between posterior cortical activity and IIT was an “auxiliary prediction” of the theory – resting on some strong and difficult to establish assumptions about anatomy. Some bridge between the theory and the human brain was clearly necessary to get the empirical work off the ground. But if we remove the gloss of GWT vs. IIT, then the fact that the back of the brain shows strong decoding of visual stimulus features is of no surprise at all to neuroscience, and should hardly shift a reasonable person’s credence in IIT. And remember that this was only the first of two major experiments, with the second yet to be reported.
As an aside, I have broader misgivings about relying on anatomy as a way of testing computational-level theories. Neural populations show a complex relationship to computation, and it’s certainly not as simple as “one region (or set of regions), one theory”. This sharply contrasts against Doris Tsao’s talk and the general cutting edge of mechanistic neuroscience. I don’t think Tsao cares too much about where in the brain her face-space neurons are. Instead, what matters is the computation – the remarkable fact that neural populations seem to be doing linear algebra in a 50-parameter face space. Localizationist neuroimaging and lesion studies are still critical in providing clues about the broader organisation of the system, and highlighting where to dig deeper – compare the classical discovery of the fusiform face area by Nancy Kanwisher and colleagues in the 1990s. But “where to look” is an intermediate step –it doesn’t test a theory. I suspect that the most impactful aspect of the Cogitate project will be in providing the kind of rich data (especially intracranial data) that will allow such computational hypotheses to be tested in detail.
So what is the upshot here? First of all, I came away from NYC clear that consciousness science is in excellent health, with considerable untapped potential. This potential will be realised by cultivating a humble focus on elements of a working model of subjective experience that can be revised or upgraded as the data come in. This endeavour is going to be significantly accelerated by the kind of big team-science projects pioneered by the Cogitate team. But in the messy world of biology, we should expect an adversarial collaboration experiment to almost always lead to incremental progress, rather than outright winners and losers. We are moving beyond the era of winner-take-all grand sweeping theories of consciousness. Just don’t believe everything you read in the papers.
*In a previous version of this post I had erroneously linked the headline in The Economist, “Thousands of species of animals probably have consciousness” to the outcome of the adversarial collaboration and the panpsychist implications of IIT. Instead, as Dave Chalmers pointed out on Twitter, this headline actually refers to the section at the end of the article where the presentations at the main ASSC meeting on animal consciousness are covered, and particularly to Oryan Zacks’ talk on vertebrate anatomy and GWT.
The Elusive Self 