Ambitious theories of consciousness are not "scientific misinformation"
In defense of Integrated Information Theory
Yesterday over one hundred scientists, many prominent or even world-famous, debuted a signed letter [1] declaring that one of the most popular scientific theories of consciousness is “pseudoscience.” The letter is directed at Integrated Information Theory (IIT) [2], and its signers go so far as to say that:
As researchers, we have a duty to protect the public from scientific misinformation.
Sparked by the recent press around an international “adversarial collaboration” that pitted the predictions of theories of consciousness against one another, the letter is quite short, just a few paragraphs, and the number of signees outweighs the number of citations by an order of magnitude. Yet the names include many well-known researchers in the field, such as Hakwan Lau, Joseph LeDoux, Bernard Baars, Patricia Churchland, Daniel Dennett, and Keith Frankish, along with plenty more (although there are noticeable absences).
I am not a fan of this letter. Everyone who signed it acted irresponsibly. Why? There’s an issue beyond its specific content. I’ve been saying for years that as a fledging science consciousness research should worry about hanging out too much dirty laundry. If too much is hung out, then petty infighting can destroy an already fragile field. My greatest fear is that we get another “consciousness winter” wherein just talking about consciousness is considered pseudoscientific bunk. This was the state of affairs throughout most of the 20th century, and it set neuroscience back decades.
I have privately expressed the worry of a consciousness winter, especially one instigated by professional jealousies and rivalries, to multiple people who signed this letter. Every outside critic arguing that consciousness shouldn’t be studied (a bad intellectual position, but a popular one) will latch onto this letter. I have no problem with dramatizing consciousness research, nor bringing its debates to the public—heck, I wrote an entire novel about it!—but that’s because I think the debates are interesting and noteworthy and attractive, rather than things to be decided by a big panel of experts declaring certain ideas “scientific misinformation.”
The letter is so bad that I’m forced to reply to it and defend IIT. Which is surprising because I just published an entire book in which several chapters track IIT’s failures and limits [3]. I’ve also published papers that explore IIT’s problems—in fact, I’ve explored the main issue referenced in the letter in greater technical detail [4] than anyone else in the world (my work is not cited). I got my PhD working with Giulio Tononi, IIT’s originator, and I’m more familiar with it than most of the people who signed this letter. So I’m going to go through the letter in its entirety and refute every point, demonstrating why this never should have been written.
(Disclaimer: despite my personal connection to IIT, I don’t get anything out of writing a defense of IIT. I don’t help develop the theory anymore, and I haven’t published with Giulio, communicated with him, or even asked him for a recommendation letter, in many years.)
The letter begins with:
The media, including news articles in both Nature and Science, have recently celebrated the Integrated Information Theory (IIT) as a ‘leading’ and empirically tested theory of consciousness. We are writing as researchers with some relevant expertise to express our concerns. The media coverage sprang from a public event where the authors of a large-scale adversarial collaboration shared their findings, which were reported as empirically testing and partially supporting IIT.
Given that IIT is one of the most popular theories of consciousness, and the IIT literature has been cited thousands of times, it can appropriately be called “leading” by the media. There are a few other “leading” theories as well, like Global Neuronal Workspace Theory (GNWT) or perhaps Higher-Order Thought (HOT) theory. In other words, we simply don’t know how consciousness works, and there is a set of competing scientific ideas, and some are very popular.
Just like the other theories, IIT has sparked a lot of empirical work, like on predicting [5] which patients will recover from a coma. Does this extensive literature [2] count as “empirically tested?” Well, depends on what you mean. No study has empirically confirmed IIT. . . but what the news media were absolutely not doing was reporting that IIT had been “empirically confirmed” by the adversarial collaboration! In fact, none of the linked news stories the letter cites as “scientific misinformation” come close to claiming that IIT in its entirety has been empirically confirmed, in any terminology. Here’s one of the sources the letter cites, Science [6], giving an explanation that was typical in the media coverage:
The results were surprisingly mixed. When it came to decoding different categories of objects, the data provided strong support for GNWT (Global Neuronal Workspace Theory). But when it came to decoding the orientation of faces, IIT was the better fit.
Or here’s The Economist, which the letter also cites as a source of “scientific misinformation” about IIT.
In the event, those hoping for a definitive victory were disappointed, though IIT seemed to have won on points. (Some data remain to be analysed, so this judgment may yet be strengthened or weakened.) But what everyone did agree was that no clear neural correlates of consciousness had been seen.
Another of the five sources cited by the letter, The New York Times [7], also described the results as “mixed.” Did anyone who signed the letter read any of the marshaled examples of “scientific misinformation” in the news media? Because all five of the cited news articles prominently featured ambiguous statements like the above examples. All five. Zero of them say, or imply, anything else. The entire motivation of the letter is just flat-out wrong.
And it’s worth pausing at this. “Scientific misinformation” has become a loaded word. It’s been used to describe the lab leak theory of Covid back when it was banned, as well as climate change denial. Using the politically-charged “scientific misinformation” term to describe the reasonable articles that the letter cites is—and I’m not going to mince words here—Machiavellian. Unless merely saying anything about IIT should be forbidden, which is what the claim of the letter appears to be.
The letter continues.
This message was communicated directly to journalists and the general public prior to the preprint being available, and hence, prior to peer-review.
This may sound nefarious, but a press release going out about the completion of a big project “prior to peer-review” is not concerning. I can’t say for certain, but it was likely authored by the Templeton organization itself, who funded the adversarial collaboration to the tune of 20 million dollars. It’s unsurprising that the Templeton organization promoted it to media, and pretending this doesn’t happen regularly within science is painting a false picture of some sort of evil scheme.
The experiments seem very skillfully executed by a large group of trainees across different labs.
They do seem that way.
However, by design the studies only tested some idiosyncratic predictions made by certain theorists, which are not really logically related to the core ideas of IIT, as one of the authors himself also acknowledges.
The idea that somehow the predictions associated with GNWT were direct tests of the theory, and meanwhile the predictions associated with IIT were “not logically related to the core ideas,” is simply not true. Here’s The New York Times describing the predictions:
The Global Workspace Theory predicted that the prefrontal cortex would send out only short bursts of information — one when a picture first appeared, and then another when it disappeared. But the Integrated Information Theory predicted that the back of the brain would be continually active throughout the time that volunteers perceived an object.
Those predictions are both overly broad caricatures of the theories. They are “logically related” to the theories, obviously, but they are not direct detailed tests of them as a whole. A true criticism is that the predictions are underdetermined by the theories, in that they could be taken as supporting evidence for any number of alternative theories. But that has nothing to do with IIT. The real issue is the paucity of neuroimaging. Pretending that other theories of consciousness are somehow directly testable in their entirety, and offer neuroimaging results unique to those theories, is just untrue.
The findings therefore do not support the claims that [IIT] itself was actually meaningfully tested, or that it holds a ‘dominant’, ‘well-established’, or ‘leading’ status. This important nuance was unfortunately lost in the media coverage.
Actually, none of the cited news articles used the term “well-established” or “dominant,” to describe IIT, and all of them were nuanced. Two cited news articles did use the term “leading” but they used it to describe both IIT and GNWT.
These claims of dominance have also been questioned in the scientific community, yet they have been repeatedly broadcast to the public by proponents of IIT over the years.
Yet again, the word “dominance” doesn’t appear in any of the cited news articles. And while one could make the argument that IIT gets more press coverage than it should, you could say this about many areas of science: maybe string theory got more coverage than it should have, maybe the “gene’s eye view” of evolution got more coverage than it should. Etc. Perhaps the interest in IIT has a reason? It’s arguably far more ambitious in what it seeks to explain, so is it any surprise it is, e.g., mentioned on social media more than other theories? They even admit as such in the next section:
IIT is an ambitious theory, but some scientists have labeled it as pseudoscience. According to IIT, an inactive grid of connected logic gates that are not performing any useful computation can be conscious—possibly even more so than humans; organoids created out of petri-dishes, as well as human fetuses at very early stages of development, are likely conscious according to the theory; on some interpretations, even plants may be conscious. These claims have been widely considered untestable, unscientific, ‘magicalist’, or a ‘departure from science as we know it.’
So if a scientific theory of consciousness has counterintuive predictions it is necessarily pseudoscience? Cerebral organoids are bits of cloned human brains grown in petri dishes—saying they might have consciousness is not wild at all! In fact, regulatory agencies have strongly considered the possibility. The idea that plants might be conscious is not popular, but it is definitely not untestable, unscientific, or “magicalist” (not a word). The idea that early-stage fetuses might have some sort of stream of consciousness is imaginable to, well, a lot of people frankly, and thus all the political debate.
The only example that might be truly counterintuitive is the “inactive grid of connected logic gates” being conscious. But it’s worth noting that IIT does not claim such a grid of logic gates would have an interesting consciousness—it would be more like an abstract bare spatial awareness and nothing else. And the particulars depend on the example and the version of IIT: IIT has gone through four different main versions so far as the theory has been refined, and later versions give different predictions for cases like this.
The letter’s claim that IIT is “untestable” ties into the next part, the only actual argument in the whole text:
Given its panpsychist commitments, until the theory as a whole—not just some hand-picked auxiliary components trivially shared by many others or already known to be true—is empirically testable, we feel that the pseudoscience label should indeed apply. Regrettably, given the recent events and heightened public interest, it has become especially necessary to rectify this matter.
First, IIT is not really “panpsychist” in the classic sense, which John Searle described as when consciousness is “spread like jam” over the universe. Most things are not conscious in the theory, just sets of interactive mechanisms (like neurons) with the greatest degree of integrated information (like brains). Consciousness, according to IIT, might be more wide-spread than we think, but it is neither universal nor arbitrary.
Regardless, the key point they are making is that:
until the theory as a whole. . . is empirically testable, we feel that the pseudoscience label should indeed apply.
So, what does the letter mean by saying that the theory “as a whole” must be empirically testable? Because as I’ve pointed out, no experiments have tested, say, uniquely GNWT as a whole. So to deduce their meaning we must examine the citations given in this section, which concern two particular issues with IIT.
The first citation given concerns the “uniqueness problem,” [8] which emphasizes a result long known by those of us who worked on IIT: that in systems that have mathematically perfectly symmetric interactions (the specifics are complicated, but just imagine this like a geometric shape), IIT gives conflicting answers as to the boundaries of the consciousness. The answer IIT gives is sort of like: mu. Both. Neither. Unask the question. Does this ambiguity disprove IIT? Well, perfect mathematical symmetry doesn’t exist in actual nature. Nothing is truly symmetric down to the atom. So it wouldn’t be a problem in practice. And it’s not even really a problem in theory either: there are plenty of examples in science where something isn’t really X or Y, but rather some sort of weird violation of the excluded middle. Where’s the electron, exactly, before it’s measured? Mu. Unask the question.
I’m not saying that you have to be convinced IIT gives the right answer in such cases. I’m saying that just because some edge cases that can’t exist in our physical universe because they involve perfect mathematical symmetry receive a koan-like “the consciousness is both X and Y” from the IIT algorithm doesn’t render IIT a pseudoscience.
The second citation given is called “the unfolding argument” [9]. Basically, imagine that you measure the integrated information of a complicated neural network (like one with lots of parts and interactions) and it’s high. Since the input/output behavior of all neural networks can be approximated by a large enough feedforward neural network, you can then “unfold” that complicated neural network into some functional equivalent (it has the same input/output behavior) feedforward neural network. For such networks, the integrated information is low. So which is correct? Is it high? Or low? The input/ouput is the same in both cases.
This is indeed an interesting argument, although the paper itself was a bit confused in how it was presented. Johannes Kleiner and I quickly did a follow-up that abstracted and generalized the issue into what we called the broader “substitution argument” [4], identifying the real problem: that the inferences about consciousness an experimenter makes that are based on behavior (like someone’s report about consciousness) can be varied independently from a theory’s predictions about consciousness (like those based on neuroimaging). We also showed that this independence assumption applies to all contemporary theories of consciousness we know of!
Now, this does seem a substantial problem for theories of consciousness. That’s why I wrote the paper. However, there are also possible solutions to the substitution argument, some more satisfying (like moving past contemporary theories) than others (like appeals to elegance, or only allowing falsification in already known-to-be conscious systems, like human brains). Ultimately, a theory of consciousness may have fundamental limits and look different in its final form compared to other scientific theories. I just published an entire book arguing precisely this!
But what I’d stress to its critics is that IIT’s problems are your problems. That is, most problems supposedly unique to IIT are not, it’s just that IIT is more ambitious and well-formalized, so it actually tackles head-on the difficulties of a theory of consciousness, and you notice the problems and blame IIT in particular.
What do I mean by “well-formalized?” I mean that IIT can act like an algorithm wherein it can be “fed” any system, anything at all—your brain, your laptop, an alien’s brain, the sun, whatever—and the theory can tell you if the thing is conscious or not, and to what degree, and what the contents are. Unlike what some may suggest, often the answer is “This thing isn’t conscious at all.” Now, does being well-formalized mean it’s correct? No. It just means IIT always gives you an answer.
No other theory of consciousness can do that. This is why IIT is a good model theory. If any of the other theories were to ever be as well-formalized as IIT, then they would inherit many of its problems—which is precisely what Kleiner and I showed. IIT being so well-formalized is what makes it possible to game the measure. The thing is, with other theories of consciousness, they aren’t immune to this, it’s just that you can’t “game the measure” because there is no measure to game! The second there is one, and someone introduces a well-formalized version of Global Workspace Theory that actually distinguishes precisely what the necessary and sufficient conditions for having a “workspace” are (and therefore having consciousness are), someone else can introduce some Rube Goldberg device that satisfies those conditions and is “super conscious” (like some ridiculously large mechanistic bulletin board) and therefore give a result that strikes us as deeply counterintuitive.
That’s why you shouldn’t use the term “pseudoscience” to attack your scientific opponents (why one even has “scientific opponents” in the first place is a good question too). Really, you’re attacking yourselves. You’re just unaware of it.
The short letter—and I’ve addressed every word so far—culminates in this politically-charged paragraph:
If IIT is either proven or perceived by the public as such, it will not only have a direct impact on clinical practice concerning coma patients, but also a wide array of ethical issues ranging from current debates on AI sentience and its regulation, to stem cell research, animal and organoid testing, and abortion. Our consensus is not that IIT and its variants decidedly lack intellectual merit. But with so much at stake, it is essential to provide a fair and truthful perspective on the status of the theory. As researchers, we have a duty to protect the public from scientific misinformation.
Are the signees actually endorsing what this sounds like: that a theory of consciousness concluding early-stage fetuses might have some degree of consciousness should be considered scientific misinformation due to political considerations? If they aren’t endorsing this, I’d like to know, because that’s what it reads as.
What does this letter hope to accomplish? It’s like it comes from some alternative reality in which IIT is taken as gospel, all the media celebrate it instead of presenting it along with other theories, and all its specific problems—many of which are indeed quite real, interesting, and informative—are entirely unique to the theory (they’re not).
Ultimately, throwing the word “pseudoscience” around at the only popular theory of consciousness ambitious enough to allow for real criticism of its details advocates for only one result: to bury scientific theories of consciousness for another hundred years. This letter is a historical step backwards, striking directly at the new and hard-won legitimacy that allows consciousness to be tackled with ambitious theories, even if most of those attempts will admittedly fail. That sort of destructive legacy is not worth wanting, and I don’t think this letter in its current form—which is right now just a pre-print—is worth pursuing through to publication for the majority of the signees.
TIP occasionally gets outside sponsors which allow me to freely write and research while also keeping the results open to all. This post is sponsored by Prophetic, a company researching consciousness and neurotechnology. It is co-posted on their blog under my name. The content reflects only my own views and was entirely my creation.
References
1. Fleming, S., Frith, C., Goodale, M., Lau, H., LeDoux, J. E., Lee, A. L., ... & Slagter, H. A. (2023). The integrated information theory of consciousness as pseudoscience.
2. Tononi, G., Boly, M., Massimini, M., & Koch, C. (2016). Integrated information theory: from consciousness to its physical substrate. Nature Reviews Neuroscience, 17(7), 450-461.
3. Hoel, Erik. The world behind the world: Consciousness, Free Will, and the Limits of Science. Simon and Schuster, 2023.
4. Kleiner, Johannes, and Erik Hoel. "Falsification and consciousness." Neuroscience of Consciousness 2021, no. 1 (2021): niab001.
5. Sinitsyn, Dmitry O., Alexandra G. Poydasheva, Ilya S. Bakulin, Liudmila A. Legostaeva, Elizaveta G. Iazeva, Dmitry V. Sergeev, Anastasia N. Sergeeva et al. "Detecting the potential for consciousness in unresponsive patients using the perturbational complexity index." Brain Sciences 10, no. 12 (2020): 917.
6. Finkel, Elizabeth. "Adversarial’search for neural basis of consciousness yields first results." Science 380 (2023): 1309-1310.
7. Zimmer, Carl. "Leading Theories of Consciousness Square Off." The New York Times (2023).
8. Hanson, Jake R., and Sara I. Walker. "On the non-uniqueness problem in integrated information theory." Neuroscience of Consciousness 2023, no. 1 (2023): niad014.
9. Doerig, Adrien, Aaron Schurger, Kathryn Hess, and Michael H. Herzog. "The unfolding argument: Why IIT and other causal structure theories cannot explain consciousness." Consciousness and cognition 72 (2019): 49-59.
I find the letter baffling. IIT is fundamentally a bunch of math. The math may be wrong or it may map inappropriately into the empirical world, but declaring it incorrect (which underlies a legitimate question) is entirely different from declaring it pseudoscience, which underlies a subjective judgment fraught with political and other types of debatable perspectives. To me, the critical point is that, by such standards, if someone tries hard enough, any theory can be argued to be pseudoscience. It smacks of prejudice. At least some of the theorists signing the letter should know better.
Looks like everyone is learning to play the same narrative game: present your group as the serious levelheaded consensus; denounce your enemies as misinformation or some hate crime; demand that media gatekeepers follow your recommendation and censor or disqualify your enemies.