Setting the record straight with Robin Carhart-Harris and Enzo Tagliazucchi

Clouds on the horizon shall not obscure the light of facts.
(Photo by Bernardo Kastrup, hereby released into the public domain.)

As many of you know, I have been recently embroiled in a public exchange with a number of neuroscientists involved in the study of psychedelics. The exchange is about what I believe to be misinterpretations and misrepresentations of their results, as Prof. Edward Kelly and I argued in an opinion piece on Scientific American. Two days ago, the researchers replied on that same forum. I have commented on and rebutted their reply in this blog.

Yesterday, one of the researchers—Enzo Tagliazucchi—engaged in a de facto Twitter debate with me. Those interested can go on Twitter to peruse the many threads of that conversation. Here, however, I want to focus on a particular point, which I feel forced to vigorously counter. You will soon understand why.

First, a little background: about four years ago, one of the researchers, Robin Carhart-Harris, was busy promoting the results of the then latest psychedelic study of his team. This included a piece he wrote for The Conversation and an interview he gave to the Washington Post, amongst a number of other media engagements. Upon comparing his media statements with the actual study in question, however, I realized that things didn't add up: what the study actually said did not—either directly or by implication—support what Carhart-Harris was telling to the media.

The issue here seemed to be a basic misunderstanding of the signal processing analysis carried out in the study. Carhart-Harris was conflating changes in BOLD—a proxy for brain activity—level with changes in BOLD variance. I even went on to explain this difference in a 2014 post in this blog. I emailed Carhart-Harris to try and clarify the situation. Carhart-Harris then added Tagliazucchi to the discussion and, in a 2014 email to me, Tagliazucchi confirmed that I was correct.

But in my Twitter exchange with Tagliazucchi, I was accused of having been the party who misunderstood the issue in the first place! Here are the relevant tweets:

brain activation does not correspond to oscillatory activity! please survey the literature on the roles of these oscillations, especially the alpha rhythm.
We had a long time ago the same discussion when you confused changes in BOLD activity levels with changes in BOLD variance

— Enzo Tagliazucchi (@ETagliazucchi) October 27, 2018

The initial accusation of my having misunderstood the relationship between oscillatory activity and brain activation did not bear out upon further debate, as you can see if you peruse the broader Twitter exchanges. The salient issue here, however, is the accusation that—precisely opposite to fact—I was the party that conflated BOLD level with BOLD variance: in actuality, I was the one who noticed Carhart-Harris' conflation of the two and sought the authors out via private email to alert them to it.

This is amazing! CARHART-HARRIS confused BOLD activity levels with BOLD variance, and I corrected him. You know it because you confirmed it to me by email. You and Robin asked me not to release those emails, which I've respected for over four years. But if needed we can clarify.

— Bernardo Kastrup (@BernardoKastrup) October 27, 2018

Carhart-Harris had asked, after the fact, that I not make our email exchange public. I have honored this request for about four years now. Yet, here is Tagliazucchi publicly claiming the opposite of what actually happened in those emails, portraying me as someone confused with the neuroscience in question and in need of redressing. This puts me in a very uncomfortable position.

He proceeded to double-down on his claim and dare me:

I recall a different story - me explaining that the mean activity of the signal does not inform on the variance and vice-versa. Same goes for “brain activity“ and “brain oscillations“. But feel free to dig into your inbox.

— Enzo Tagliazucchi (@ETagliazucchi) October 27, 2018

For clarity, I don't doubt that he was being honest in these tweets, in the sense that he truly recalls the story in a distorted manner. Admittedly, the actual story is somewhat surprising: a philosopher, with no formal background in neuroscience, suddenly emailing the authors of a neuroscience study to correct their interpretation of their own study? The opposite would be more expectable, of course. Nonetheless, facts are facts. So I accepted the challenge:

OK, can I make it all public? I can do it this weekend, no problem. Are you okay with it?

— Bernardo Kastrup (@BernardoKastrup) October 27, 2018

If you are going to start reproducing correspondence we had in the past, it would imply an effort on my part checking that there is no bias in what you chose to reproduce and frankly I don't have the time, but as long as you keep it honest, there is nothing to hide from my part

— Enzo Tagliazucchi (@ETagliazucchi) October 27, 2018

I will edit nothing. I will just do a verbatim dump. But it does disappoint me beyond measure that you can INVERT the story! I reached out to you and Robin to alert you that higher spectral power did NOT imply a higher-amplitude time-domain signal, since you discard phase info.

— Bernardo Kastrup (@BernardoKastrup) October 27, 2018

Let me be absolutely honest and clear: I had agreed to Carhart-Harris' after-the-fact request not to make the emails public. I stick to my word and the very idea of disclosing those emails gives me a wrenching feeling in my gut. I don't take this at all lightly. But put yourself in my position now:

1. One of the people involved in that email exchange is now publicly claiming—without maliciousness, as I believe the case to be, but nonetheless publicly—that what happened was the opposite of what actually did. This inversion of the truth publicly portrays me as the ignorant and confused party, instead of the party that found the misunderstanding and then informed the authors about it. I have the emails to prove it. Should I continue to sit on them while Tagliazucchi's incorrect tweet stays online for all to see?
2. These same people—Carhart-Harris and Tagliazucchi are both co-authors of the Scientific American reply attacking me and Prof. Kelly—are now publicly accusing me of misrepresenting their psychedelic studies. I have the emails to prove that Carhart-Harris himself has been—and continues to be, insofar as he has never amended his erroneous earlier statements, as I shall discuss below—the chief misrepresenter of the work. Should I continue to sit on them?
3. In my Twitter exchange with him, Tagliazucchi repeatedly accused me of being ignorant of the basics of neuroscience. Consider this tweet, for instance:
   

   It is not a point about “silly ad-hominems“ - I would respond even if there were. But it is very difficult to discuss with someone who needs to revisit Neuroscience 101. The mere fact that he identifies “alpha power“ with “brain activity“ makes it so hard to engage.

   — Enzo Tagliazucchi (@ETagliazucchi) October 27, 2018

   I address the false accusation that I "identify 'alpha power' with 'brain activity'" in other tweets of the exchange. But be that as it may, should I continue to sit on an email exchange that demonstrates that, although some 101 basics were indeed misunderstood, it was not me who misunderstood them?
4. Tagliazucchi, one of the people involved in the original 2014 email exchange, has now given me public permission to disclose those emails. Should I continue to sit on them?

I feel Carhart-Harris and Tagliazucchi have left me no alternative but to defend myself and disclose the emails, for the reasons above. The record must be set straight. I can't remain silent in face of this onslaught.

(Post-publication clarification: In private correspondence after the original publication of this post, Tagliazucchi told me that his intent was to give me permission to publish only his own messages in the three-way exchange of 2014, not the messages he was commenting on. I acknowledge the clarification. But Tagliazucchi's messages alone would lack all context and much of the substance they refer to, thereby making it impossible for me to defend myself against his claims in point 1 above. Let it be clear, nonetheless, that I have four reasons to publish the 2014 exchange, only one of which is Tagliazucchi's permission. It has also been claimed, in private correspondence to me, that I am publishing messages originally shared with me in confidence. This, however, is not true. The messages were originally exchanged freely and openly, without any prior understanding regarding confidentiality, either explicit or implicit. I simply emailed Carhart-Harris asking for clarifications regarding public articles of his. There was no reason to think of confidentiality or for him to treat me any differently than any member of the public; he didn't know me back then. Only thereafter did Carhart-Harris ask me to not divulge the exchange. More specifically, this happened after he read the draft of an essay I sent to him, which was eventually published in my book Brief Peeks Beyond. In that draft, it was clear that my interest in Carhart-Harris' work was motivated by a metaphysical position I was trying to argue for. I shared the draft with him because I wanted to clarify in it, by reference to passages of our email exchange, that there was actually no finding of increased brain activity in the technical paper in question, despite how it was portrayed in the media. Carhart-Harris did not agree to it. My having not disclosed this material since then was a gesture of respect and courtesy towards Carhart-Harris and Tagliazucchi. But since Tagliazucchi, out of the blue and without prompting by me, now chose to refer to that original 2014 email exchange and characterize it as the opposite of what it actually was, the basis for that courtesy is nullified and I feel effectively forced to publish the exchange.)

Below, I reproduce the 2014 emails verbatim. I didn't correct even the typos, so I apologize if the text looks sloppy at times.

Here is my original email to Carhart-Harris, dated Monday, 17 November 2014, sent at 10:52am Central European Time:

Dear Robin,
I've been following your studies with psilocybin with interest since the preliminary publications in 2011. I've been reading mostly your technical publications but, today, read a popular science article you wrote here:
http://myscienceacademy.org/2014/07/05/magic-mushrooms-expand-your-mind-and-amplify-your-brains-dreaming-areas-heres-how/
In your 2012 PNAS paper you explicitly say that psilocybin decreases brain activity mostly in the DMN and doesn't increase it anywhere in the brain. In your new HBM study you talk of an increase in variability and spectral power of activity in dream-associated areas. Naturally, an increase in variability is not necessarily an increase in activity. Similarly, an increase in spectral density is also not necessarily an increase in activity, since phase information is ignored. I've concluded then that your new study in no way contradicts your earlier findings: psilocybin has NOT been found to increase sheer brain activity (metabolism, BOLD signal) in dream-associated areas, even though the media seems to have described the study that way. I've attributed the inaccuracy to journalists.
Yet, in your own myscienceacademy writeup you wrote "that psilocybin increased the amplitude (or “volume”) of activity in regions of the brain that are reliably activated during dream sleep and form part of the brain’s ancient emotion system." You also wrote of "the principle that the psychedelic state rests on disorganised activity in the ego system permitting disinhibited activity in the emotion system." Both statements are at least highly suggestive of a direct increase in brain activity ("amplitude," "volume," "reduced disinhibition"), even though no indication of this seems to be found in your technical papers (the HBM paper, for instances, talks of an increase in amplitude of variations).
I wonder if you could help me understand the discrepancy. Have you ever found that psilocybin increases sheer brain activity (BOLD signal, metabolism) anywhere in the brain?
Kind regards, Bernardo.

Carhart-Harris replied kindly and attentively on Tuesday, 18 November 2014, in an email I received at 12:12pm Central European Time:

Dear Bernardo,
Thanks for your email. The questions you raise are entirely valid and I would summarise it thus: the term 'activity' is generic and non-specific and not really that helpful or informative. With fMRI and FDG PET do not measure neural activity directly but instead via proxies, e.g. cerebral blood flow or glucose metabolism to index 'activity'. However, 'activity' used in this way is vague and we should probably not use it and instead just refer to the proxies. It is not certain for example that ASL fMRI and FDG PET measure the same 'activity'. My preference would be to move away from any reference to absolute increases or decreases in activity in relation to the action of psychedelic drugs. We get more information in the spontaneous oscillations and fluctuations in neural activity than we do in any absolute increases or drops in cerebral blood flow for example. If we observe increases in signal variance, this means the spontaneous signal fluctuations have a higher amplitude. You could interpret this as an increase in 'activity' but like I said, I'd prefer to avoid this and instead use a more specific term, i.e. an increase in signal variance or signal amplitude.
In short, we have mostly seen decreases in our measures of neural activity when looking at the brain effects of psychedelic drugs but it would be too simplistic to say all of these measures measure something that we can generically call 'activity' - as if it is something that is absolute, a quantitative thing that rises or falls. Instead, we need to think of these signals as dynamic and then think how best to describe the specific measures. My preference is to move towards referring to increases or decreases in the 'order' or 'organisation' of the systems from which the signal is recorded. I've attached an article that may help. When trying to communicate the results in the simplest terms, we often succumb to references to increases or decreases in activity but as I've said above, this isn't really the full picture and so, it can be misleading. It's a difficult trade off, knowing when to be specific but risk people not understanding what you're talking about or being generic but then over-simplifying the matter. If ever I make the 'mistake' of referring to increases or decreases in activity (which I do) then I need to hold my hands up and say yes, this is an over-simplification. You may find evidence of this over-simplification in what I have just sent you.
With best wishes,
Robin.

Kind as it is, this answer is evasive. It renders it virtually impossible to pin down what Carhart-Harris means whenever he uses the term 'activity.' How can it sometimes denote short-term fluctuations of metabolism and other times (such as in association with their 2012 PNAS paper) metabolism itself? Why use the term 'activity' repeatedly towards the media (e.g.: "psilocybin increased the amplitude (or “volume”) of activity in regions of the brain," and "You’re seeing these areas getting louder, and more active," etc.) if one means by it different things at different times? What are we to make of it, as readers? The need to simplify things towards the general public is acknowledged, but the simplification should still preserve internal consistency and comply to the standard meaning of terms used without further definition, lest the communications become at least misleading.

In any case, comparing brain activity levels between placebo and psychedelic conditions must somehow involve a time-averaged mean amplitude of the signal for each respective condition, not merely the signal's short-term fluctuations within a condition. But an analysis of variance says nothing about the signal's time-averaged mean amplitude. I therefore insisted on my point, in an email to Carhart-Harris and his supervisor, Prof. David Nutt, dated Tuesday, 18 November 2014, 1:10pm Central European Time:

Dear Robin and Prof. Nutt,
Thank you both for your prompt and thoughtful replies, which I much appreciate. Without meaning to abuse, I have a couple of very brief comments.

>> We get more information in the spontaneous oscillations and fluctuations in neural activity than we do in any absolute increases or drops in cerebral blood flow for example. <<

This is paradigmatically very significant and has profound implications as far as the biological basis of consciousness, as I am sure you realize.

>> If we observe increases in signal variance, this means the spontaneous signal fluctuations have a higher amplitude. You could interpret this as an increase in 'activity' <<

An increase in variance (and spectral power) can be tied to higher-frequency fluctuations, not only fluctuations of higher amplitude. An increase in variance can lead to signals of lower amplitude, if the frequency components are not in phase and thus interfere destructively. Isn't this correct? Assuming it is, I will be as bold as to disagree with you here. I acknowledge the ambiguity behind proxy measurements of activity, but activity itself is a clean concept tied to metabolism. Variations of levels of metabolism are not activity; metabolism is. Personally, I find it confusing to conflate these terms.

>> I'd prefer to avoid this and instead use a more specific term, i.e. an increase in signal variance or signal amplitude. <<

I concur. I have a brief question: in the HBM paper you report increases in the amplitude of fluctuations of the BOLD signal (i.e. amplitude of the delta in time). This leaves it unclear whether the amplitude of the BOLD signal itself, during bursts, was higher than in placebo controls. I could imagine that, even though time-averaged BOLD is not higher in the psychedelic state, brief bursts could be. Did you observe higher-amplitude bursts of the BOLD signal in the psychedelic state as compared to placebo controls?
On a more personal note, Robin, I enormously appreciate the work you and Prof. Nutt have been doing. You are pioneers in the full sense of the word. I am keenly aware of the pressures you are probably under, given the sensitivity of your work for our scientific and philosophical worldviews in general, way beyond neuroscience. Though I am critical of the ambiguity of your recent communications, I am very understanding of the potential motivations behind it, and tend to believe I would not have done better than you are doing.
Sincere regards,
Bernardo.

Carhart-Harris replied very openly and honestly in an email dated Tuesday, 18 November 2014, which I received at 2:09pm Central European Time:

Dear Bernado,
I've copied in Enzo, first author of the HBM paper to address your questions re signal variance. My understanding is that an increase in signal variance as it was measured in the HBM paper is equivalent to an increase in signal amplitude. Enzo may wish to comment on this.
As for reproducing my email response in print. If it's ok, I'd rather it not be. It was written somewhat in haste and I'd consider this more free communication rather than something sufficiently considered for publication. Perhaps if I could review and edit however, it could it principle be ok.
As for the issue re bursts. The increase in signal variance applies for a 5 min continuous period of scanning (post-infusion) and so is time-averaged. We did not specifically look at periodic bursts but this would be something interesting to look at. Would this be a Wavelet-like analysis Enzo? We'd need to know how to define a 'burst' in a systematic way. Since scanning was pure 'rest' there are no behavioural 'tags' to time lock 'bursts' to - but this is not to say they do not occur or that they are not behaviourally relevant. We'd have to be careful not to interpret motion artefact as a neurally generated burst. We have looked at 1 min time bins (see entropy analysis in HBM paper) Enzo is an expert on these more dynamic measures.
Best wishes,
Robin.

Carhart-Harris admits here that his "understanding is that an increase in signal variance as it was measured in the HBM paper is equivalent to an increase in signal amplitude." This is not true. Signal variance is equivalent only to the RMS (not peak) amplitude of the fluctuations of the signal; that is, what is left of the signal after one first subtracts the signal's mean value from it. Performing this subtraction, however, amounts to discarding the core information about the difference in time-averaged brain activity levels between placebo and psychedelic conditions, which can't then be compared. There seems to be, as such, a misunderstanding here on Carhart-Harris' part, which could explain the consistent media misreporting (e.g. this and this) claiming that the study in question found increases in activity in dream-related regions of the brain. The 'activity' reported to increase in these dream-related regions is, at any rate, something quite different from the activity proper reported to decrease in his 2012 PNAS paper. Under the normal denotation of 'activity' as a measure of metabolism, this 2014 paper showed no increases in activity—which, by the way, renders it consistent with the earlier 2012 results—despite Carhart-Harris' messages to the media.

Upon this email, Tagliazucchi, lead author of the study in question, replied on Tuesday, 18 November 2014, in an email I received at 2:57pm. In the email, he confirms my assessment:

Dear all,
Bernardo: indeed, variance is in this context just a measure of variability in the signal. Increased variance implies increased fluctuations of the signal, as opposed to a constant, unchanging signal. My interpretation of these fluctuations is that spontaneously occurring processes in the affected regions are engaging and disengaging more often under the psilocybin condition, so even if subjects are resting, spontaneous processes are more "active" and thus increase the variability of the signal. Following this line of thought, I'm interpreting signal variance as a sort of equivalent of BOLD signal amplitude but during rest (i.e. a measure of "something going on" you can apply during rest, as opposed to BOLD signal amplitude, for which you need a task paradigm).
Cheers,
Enzo

I find Tagliazucchi's "interpreting signal variance as a sort of equivalent of BOLD signal amplitude but during rest" misleading. Activity variability is not the same as activity proper; acceleration is not the same as speed. But never mind: the salient point in the context of this exchange is that Tagliazucchi recognized that my assessment was correct.

Misunderstandings do happen. I surely have misunderstood many things in my life and continue to do so. I don't think it can ever be avoided. My problem here is not Carhart-Harris' misunderstanding itself, but the false, inverted public portrayal of what happened in the 2014 email exchange. I also have a problem with something else: despite my explicit and repeated private and public requests for him to do so, as far as I know Carhart-Harris never rectified his mistaken media communications. As I write this, for instance, his original 2014 The Conversation essay remains unchanged.

Be that as it may, I hope this post sets the record straight once and for all.