Consciousness researcher and co-founder of the Qualia Research Institute. I blog at

Core interests span - measuring emotional valence objectively, formal models of phenomenal space and time, the importance of phenomenal binding, models of intelligence based on qualia, and neurotechnology.


Consciousness research as a cause? [asking for advice]

I don't see anything like that from QRI either, although someone can correct me if I missed it.


In Principia Qualia (p. 65-66), Mike Johnson posits:

What is happening when we talk about our qualia? 

If ‘downward causation’ isn’t real, then how are our qualia causing us to act? I suggest that we should look for solutions which describe why we have the sensory illusion of qualia having causal power, without actually adding another causal entity to the universe.

I believe this is much more feasible than it seems if we carefully examine the exact sense in which language is ‘about’ qualia. Instead of a direct representational interpretation, I offer we should instead think of language’s ‘aboutness’ as a function of systematic correlations between two things related to qualia: the brain’s logical state (i.e., connectome-level neural activity), particularly those logical states relevant to its self-model, and the brain’s microphysical state (i.e., what the quarks which constitute the brain are doing). 

In short, our brain has evolved to be able to fairly accurately report its internal computational states (since it was adaptive to be able to coordinate such states with others), and these computational states are highly correlated with the microphysical states of the substrate the brain’s computations run on (the actual source of qualia). However, these computational states and microphysical states are not identical. Thus, we would need to be open to the possibility that certain interventions could cause a change in a system’s physical substrate (which generates its qualia) without causing a change in its computational level (which generates its qualia reports). We’ve evolved toward having our qualia, and our reports about our qualia, being synchronized – but in contexts where there hasn’t been an adaptive pressure to accurately report our qualia, we shouldn’t expect these to be synchronized ‘for free’. 

The details of precisely how our reports of qualia, and our ground-truth qualia, might diverge will greatly depend on what the actual physical substrate of consciousness is.48 What is clear from this, however, is that transplanting the brain to a new substrate – e.g., emulating a human brain as software, on a traditional Von Neumann architecture computer – would likely produce qualia very different from the original, even if the high-level behavioral dynamics which generate its qualia reports were faithfully replicated. Copying qualia reports will likely not copy qualia. 

I realize this notion that we could (at least in theory) be mistaken about what qualia we report & remember having is difficult to swallow. I would just say that although it may seem far-fetched, I think it’s a necessary implication of all theories of qualia that don’t resort to anti-scientific mysticism or significantly contradict what we know of physical laws. 

Back to the question: why do we have the illusion that qualia have causal power? 

In short, I’d argue that the brain is a complex, chaotic, coalition-based dynamic system with well defined attractors and a high level of criticality (low activation energy needed to switch between attractors) that has an internal model of self-as-agent, yet can’t predict itself. And I think any conscious system with these dynamics will have the quale of free will, and have the phenomenological illusion that its qualia have causal power. 

And although it would be perfectly feasible for there to exist conscious systems which don’t have the quale of free will, it’s plausible that this quale will be relatively common across most evolved organisms. (Brembs 2011) argues that the sort of dynamical unpredictability which leads to the illusion of free will tends to be adaptive, both as a search strategy for hidden resources and as a game-theoretic advantage against predators, prey, and conspecifics: “[p]redictability can never be an evolutionarily stable strategy.”

algekalipso's Shortform

[Related to:  Logarithmic Scales of Pleasure and Pain; Anti-Tolerance Drugs]

Millions of people suffering from benzo/gabapentin/phenibut/alcohol withdrawal across the world thinking that tapering is the only solution, while it seems like people in Italy and Japan already figured out how to reverse tolerance without side-effects within a week? It's called Flumazenil and it's a GABAa  antagonist, which when taken in microdoses can up-regulate GABA receptors (hear me out: up-regulation and other tolerance mechanisms are *not* proportional to subjective effect size - this insight makes all the difference).

See the wikipedia entry on it:

In Italy, the gold standard for treatment of high-dose benzodiazepine dependency is 8–10 days of low-dose, slowly infused flumazenil.[12] One addiction treatment centre in Italy has used flumazenil to treat over 300 patients who were dependent on high doses of benzodiazepines (up to 70 times higher than conventionally prescribed) with physicians being among the clinic's most common patients.[13]

Epileptic patients who have become tolerant to the anti-seizure effects of the benzodiazepine clonazepam became seizure-free for several days after treatment with 1.5 mg of flumazenil.[14] Similarly, patients who were dependent on high doses of benzodiazepines (median dosage 333 mg diazepam-equivalent) were able to be stabilised on a low dose of clonazepam after 7–8 days of treatment with flumazenil.[15]

Flumazenil has been tested against placebo in benzo-dependent subjects. Results showed that typical benzodiazepine withdrawal effects were reversed with few to no symptoms.[16] Flumazenil was also shown to produce significantly fewer withdrawal symptoms than saline in a randomized, placebo-controlled study with benzodiazepine-dependent subjects. Additionally, relapse rates were much lower during subsequent follow-up.[17]

In vitro studies of tissue cultured cell lines have shown that chronic treatment with flumazenil enhanced the benzodiazepine binding site where such receptors have become more numerous and uncoupling/down-regulation of GABAA has been reversed.[18][19][20] After long-term exposure to benzodiazepines, GABAA receptors become down-regulated and uncoupled. Growth of new receptors and recoupling after prolonged flumazenil exposure has also been observed. It is thought this may be due to increased synthesis of receptor proteins.[21]

Flumazenil was found to be more effective than placebo in reducing feelings of hostility and aggression in patients who had been free of benzodiazepines for 4–266 weeks.[22] This may suggest a role for flumazenil in treating protracted benzodiazepine withdrawal symptoms.

 Low-dose, slow subcutaneous flumazenil administration is a safe procedure for patients withdrawing from long-term, high-dose benzodiazepine dependency.[23] It has a low risk of seizures even amongst those who have experienced convulsions when previously attempting benzodiazepine withdrawal.[24]

See also a video I made about why our common-sense view of how drug tolerance works gets in the way of actually solving this crisis.

Cost-Effectiveness of Air Purifiers against Pollution

Now that the California fires are raging, it may be time to bring up a few additional reasons why HEPA filters make a lot of sense. I don't know how much this changes the cost-benefit analysis, but I think it is important to take into account:

1) Right now the PM2.5 outside my apartment is 230. Inside it's 40. A week ago the PM2.5 was 100, and inside it was 8. By having a HEPA filter inside, I've been seeing reductions of PM2.5 between 80% and 90%. I also saw this two years ago, and it's been a rather consistent pattern.

2) The idea that non-linearity makes the benefits strictly less than linearity, and therefore that assuming linearity will lead to an optimistic assessment is questionable. In particular, I grant this is true with "diminishing returns" curves. But it's not true with S-shaped curves. So, if it is true that the negative health effects of PM2.5 are concave below 20 and convex above 20, then the assumption of linearity will lead to an underestimation of the positive health benefits of HEPA filters for places with relatively clean air.

3) As a special case of (2), I would expect that giving your lungs "time to breath" (so to speak) might be really good to let them heal, and also allow your cardiovascular system to recover from inflammation. So there may be some extra benefits to being in places that have close to 0PM2.5 for at least some periods of time. And lastly,

4) I do think that the case for massively reducing the economic cost of HEPA filters should be considered more thoroughly. If subsidized at the governmental level, how cheap could these filters become? My suspicion is that they can become extremely cheap, turning them into a utility.

Thank you for the analysis and for bringing this topic to attention of EAs (whose saved micromorts may, well, ultimately have compounding benefits for all). Cheers!

Making discussions in EA groups inclusive

Like many other problems that EAs are aware of, the particular incident you described comes from an outlier that drives the mean significantly forward (I of course know who you are talking about, and the fact that many who've been in EA for a long time know as well should indicate that this is both rare in terms of % of people yet perhaps not that rare in terms of % of drama it accounts for).

The key insight here is that the long-tail matters. As a rough prior we could anticipate that 80% of the drama will come from 20% of people (in my experience this is even more skewed, to perhaps 98% of drama coming from 2% of people). In which case, advocating for self-censorship in general in the community is stifling and unnecessary for the bulk of people (who already doubt themselves), and desperately necessary for the outliers who just march forward without much self-awareness in some or other controversial direction, as if mandated by a higher power to cause as much drama as possible.

If we recognize that the problem per person follows a long-tail distribution, our strategies should look very different than if it was a kind of normal/Gaussian distribution.

EAGxVirtual Unconference (Saturday, June 20th 2020)

Hi Aidan!

Thank you ^_^

We are collaborating with John Hopkins and Stanford researchers on a couple of studies involving the analysis of neuroimaging data of high-valence states of consciousness. Additionally, we are currently preparing two key publications for peer-reviewed journals on our core research areas.

Off the top of my head, some well-known researchers and intellectuals that are very positive about our work include: Robin Carhart-Harris, Scott Alexander, David Pearce, Steven Lehar, Daniel Ingram, etc. (e.g. Scott acknowledged that QRI put together the paradigms that contributed to Friston's integrative model of how psychedelics work before his research was published). Our track record so far has been to foreshadow by several years in advance key discoveries later proposed and accepted in mainstream academia. Given our current research findings, I expect this to continue in the years to follow.

Cheers! :)

EAGxVirtual Unconference (Saturday, June 20th 2020)

*Logarithmic Scales of Pleasure and Pain*

Recall that while some distributions (e.g. the size of the leaves of a tree) follow a Gaussian bell-shaped pattern, many others (e.g. avalanches, size of asteroids, etc.) follow a long-tail distribution. Long-tail distributions have the general property that a large fraction of the volume is accounted for by a tiny percent of instances (e.g. 80% of the snow that falls from the mountain will be the result of the top 20% largest avalanches).

Keeping long-tails in mind: based on previous research we have conducted at the Qualia Research Institute we have arrived at the tentative conclusion that the intensity of pleasure and pain follows a long-tail distribution. Why?

First, neural activity on patches of neural tissue follow log-normal distributions (an instance of a long-tail distribution).

Second, the extremes of pleasure and pain are so intense that they cannot conceivably be just the extremes of a normal distribution. This includes, on the positive end: Jhana meditation, 5-MeO-DMT peak experiences, and temporal lobe epilepsy (Dostoevsky famously saying he'd trade 10 years of his life for just a few moments of his good epileptic experiences). On the negative end, things like kidney stones, cluster headaches, fibromyalgia, and migraines top the charts of most intense pain.

And third, all of the quantitative analysis we conducted on a survey about people's best and worst experiences showed that the ratings, comparisons, and rankings of such experiences was far more consistent with a long-tail distribution than a normal distribution. The data could not be explained with a Gaussian distribution; it fit very nicely a log-normal distribution.

This is an *important*, *tractable*, and *neglected* cause.

1) Important because we may be able to reduce the world's suffering by a significant amount if we just focus on preventing the most intense forms of suffering.

2) Tractable because there are already many possible effective treatments to these disorders (such as LSD microdosing for cluster headaches, and FSM for kidney stones).

3) And neglected because most people have no clue that pain and pleasure go this high. Most utilitarian calculus so far seems to assume a normal distribution for suffering, which is very far from the empirical truth. Bentham would recoil at the lack of an exponent term when additively normalizing pain scales.

Importantly, in Effective Altruism there might be an implicit "youth" bias involved in the lack of knowledge of this phenomenon - due to the age of the people in the movement, most EA activists will not themselves have had intensely painful experiences. Thus, why it is so crucial to raise awareness about this topic in the community (it does not show up on its own). Simply put: because the logarithmic nature of pleasure and pain is *news* to most people in EA.

For more, see the original article: Logarithmic Scales of Pleasure and Pain

And a presentation about it that I shared at the New York EA chapter:


[I would prefer the late session if possible]


[June 22 2020 edit: Thank you all for attending and/or voting for this talk! I appreciated your engagement and questions! For people who would like to see the video, here it is: Effective Altruism and the Logarithmic Scales of Pleasure and Pain]

What are the key ongoing debates in EA?

Whether avoiding *extreme suffering* such as cluster headaches, migraines, kidney stones, CRPS, etc. is an important, tractable, and neglected cause. I personally think that due to the long-tails of pleasure and pain, and how cheap the interventions would be, focusing our efforts on e.g. enabling cluster headaches sufferers to access DMT would prevent *astronomical amounts of suffering* at extremely low costs.

The key bottleneck here might be people's ignorance of just *how bad* these kinds of suffering are. I recommend reading the "long-tails of pleasure and pain" article linked above to get a sense of why this is a reasonable interpretation of the situation.

Logarithmic Scales of Pleasure and Pain (@Effective Altruism NYC)

Thank you! I just left a reply to your comment. Here's a summary of the core claim:

In this account, the fact that people would naturally and spontaneously use a logarithmic scale to report their level of pain is a simple implication of the fact that you can only definitively tell that "the pain got worse" when it got 10% worse and not when it became 1 unit worse (which soon becomes hard to notice when you talk about experiences with e.g. 1000 pain units per second).
Logarithmic Scales of Pleasure and Pain: Rating, Ranking, and Comparing Peak Experiences Suggest the Existence of Long Tails for Bliss and Suffering

Hey Michael,

Thank for commenting. First of all I agree that the claims of (A) and (B) do need to be distinguished, and I admit I didn't make that conceptual distinction very clear in the article. I agree that the most important takeaway from the piece is (B), and I also think that this alone is already enough to challenge EA's prioritization methods (i.e. ultra-painful experiences are completely flying under the radar from the point of view of QALYs and similar metrics; reducing the incidence of cluster headaches, migraines, kidney stones, etc. could be an extremely cost-effective EA objective).

With that said, I would claim that (1) the article does actually provide evidence for (A), (2) taking this seriously clarifies a lot of puzzling facts about experience and how it is reported, and (3) understanding that pain and pleasure follows a long-tail (most likely a log-normal distribution) gives us a new principled way to approach cause prioritization.

I understand the fact that the pain scales of stings and cluster headaches are *by construction* logarithmic. But you have to understand that such a scale would only ever be "filled to the top" if experiences actually differed in intensity also by the same amount. The article (and presentation, which I strongly recommend you watch) explain that all of the following are consistent with the pain scales (as reported!) are actually logarithmic:

(a) the characteristic distribution of neural activity is log-normal, and under the modest assumption that intensity of experience is roughly proportional (or at least polynomially proportional) to intensity of experience, that entails the distribution of intensity is also log-normal.

(b) the above can be further understood as a kind of "neuronal weather" (see the "avalanches" metaphor in the video presentation)

(c) the predictions of the log-normal world are held by the data, and in particular:

(c1) there are few categories of experiences that capture most of the extremely good and extremely bad sensations

(c2) there is consistency in the deference judgements of the quality of experience (as seen in the deference graph), and importantly

(c3) The ratio of "1st worst or best experience vs. 2nd worst or best experience" fits a log-normal distribution and it does not fit a normal distribution.

For the above reasons, bringing up the Fechner-Weber is not, I would claim, a red-herring. Rather, I think it ties together the whole argument. Here is why:

I understand that Fechner-Weber's law maps physical intensity to subjective intensity, and that valence is not externally driven a lot of the time. But you may have missed the argument I'm making here. And that is that in one interpretation of the law, a pre-conscious process does a log transform on the intensity of the input and that by the time we are aware of it, what we become aware of are the linear differences in our experience. In the alternate interpretation of the law, which I propose, the senses (within the window of adaptation) translate the intensity of the input into an equivalent intensity of experience. And the reason *why* we can only detect multiplicative differences in the input *is because* we can only notice consciously multiplicative differences in the intensity of experience. Do you see what I am saying? In this account, the fact that people would naturally and spontaneously use a logarithmic scale to report their level of pain is a simple implication of the fact that you can only definitively tell that "the pain got worse" when it got 10% worse and not when it became 1 unit worse (which soon becomes hard to notice when you talk about experiences with e.g. 1000 pain units per second).

In other words, the scales are logarithmic because we can only notice with confidence multiplicative increments in the intensity of experience. And while this is fine and does not seem to have strong implications on the lower end of the scale, it very quickly escalates, to the point where by the time you are in 7/10 pain you live in a world with orders of magnitude more pain units per second than you did when you were in 2/10 pain.

Finally, you really need the logarithmic scales to make room for the ultra-intense levels of pleasure and pain that I highlighted in the "existence of extremes" section. If people reported their pain on a linear scale, they would quickly run into the problem that they cannot describe even something as painful as a broken bone, let along something like a cluster headache.

Load More