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).
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.
Thanks for writing this.
How would this model explain Cluster Headaches? They are not particularly more incapacitating than migraines, yet they are (possibly literally*) thousands of times more acutely painful than them. What is the role of this X1000 multiplier on phenomenal pain in such cases? As far as I can tell, in the ancestral environment nobody could have done anything to help you if you were having a Cluster Headache, and your chances of reproduction seem to be the same whether that pain was a thousand times less bad (which would still be VERY bad, but not in the level of ultra-Hellish pain). In particular, other species are known the have Cluster Headaches too, such as cats. So perhaps we should cluster pains into two buckets—those that have social significance and those that don’t. I worry that this study will make people dismiss extreme suffering in nonhuman animals, but that should only really apply to socially-useful pains. I suspect that there are many species-specific ultra-painful experiences that we will not discover (and prioritize!) unless we look for them.
That’s a good point, thank you. We should distinguish between lifetime use and current use in future surveys. Perhaps even asking whether “they worked the first time you used them” to see if people who currently use them had a better reaction to their first try relative to those who did try them at some point but do not currently use them.
I would add that other reasons why people might have used them in the past but don’t currently include “can’t access it now”, “too afraid of legal repercussions”, and “social stigma”. While discontinuing them due to side-effects and lack of effectiveness can make them look more effective than they are among the “use them” group, the other reasons for discontinuation do not have this effect. I don’t know what % of past users discontinued for which reason, and that seems like a good thing to find out.
Depends on context. In most cases the ‘we’ refers to my team and I at the Qualia Research Institute. For example: “Since a number of interviews we’ve conducted have shown that even sub-hallucinogenic doses of DMT can abort cluster headaches” refers to QRI (with other members of the research group having conducted such interviews).
I should note that the word is also used in the ‘didactic we’ sense a number of times (as in “we will explore the era of the dinosaurs together” in a National Geographic documentary).
According to “Right Concentration: A Practical Guide to the Jhanas” by L. Brasington and “The Mind Illuminated” by Culadasa, it is feasible to achieve Jhana states within two years of dedicated practice. This entails a few hours of meditation a day and attending at least one 9-day retreat over the course of this time period. The books explain in detail how to get there in a very practical and no-nonsense way.
I personally have yet to invest that time into this task, but I know that one of the other core members of the Qualia Research Institute, Romeo Stevens, is now able to achieve Jhanas thanks to his meditation practice. I do intend to do this in the near future.
Also, we are looking into doing EEG and fMRI studies on people who can enter those states as a means to test the CDNS approach to valence quantification, which is a core part of our research plan.
2019-09-04 Update: Since posting this I’ve learned about the Bradley-Terry model for obtaining latent traits based on sets of rankings (https://en.wikipedia.org/wiki/Bradley%E2%80%93Terry_model) and also that there are libraries to do this (e.g. https://pypi.org/project/choix/).
Additionally, I’ve learned about “extreme value theory”, which describes the statistical distribution of extreme values (e.g. https://en.wikipedia.org/wiki/Generalized_extreme_value_distribution) and seen some applications to other long-tail events (see: https://blog.givewell.org/2015/07/13/geomagnetic-storms-using-extreme-value-theory-to-gauge-the-risk/).
I will use those two new key statistical approaches to analyze this pilot dataset and also future iterations of this study (focused more on people who’ve experienced extremes of valence like cluster headaches or 5-MeO-DMT states). I am currently busy working on a number of other projects critical for the Qualia Research Institute, so doing this is currently on the back-burner (though of course I’m happy to hear if anyone is interested in taking on this challenge as a volunteer project).
Cheers!
To zoom in on the “logarithmic scales of pleasure and pain” angle (I’m the author), I would say that this way of seeing the world suggests that the bulk of suffering is concentrated on a small percentage of experiences. Thus, finding scaleable treatments specially for ultra-painful conditions could take care of a much larger percent of the world burden of suffering than most people would intuitively realize. I really think this should be up in the list of considerations for Cause X. Specifically:
An important pragmatic takeaway from this article is that if one is trying to select an effective career path, as a heuristic it would be good to take into account how one’s efforts would cash out in the prevention of extreme suffering (see: Hell-Index), rather than just QALYs and wellness indices that ignore the long-tail. Of particular note as promising Effective Altruist careers, we would highlight working directly to develop remedies for specific, extremely painful experiences. Finding scalable treatments for migraines, kidney stones, childbirth, cluster headaches, CRPS, and fibromyalgia may be extremely high-impact (cf. Treating Cluster Headaches and Migraines Using N,N-DMT and Other Tryptamines, Using Ibogaine to Create Friendlier Opioids, and Frequency Specific Microcurrent for Kidney-Stone Pain). More research efforts into identifying and quantifying intense suffering currently unaddressed would also be extremely helpful.
(see also the writeup of an event we hosted about possible new EA Cause Xs)
Do you think that the empirical finding that pain and suffering are distributed along a lognormal distribution (cf. Logarithmic Scales of Pleasure and Pain) has implications for how to prioritize causes? In particular, what do you say about these tentative implications:
Of particular note as promising Effective Altruist careers, we would highlight working directly to develop remedies for specific, extremely painful experiences. Finding scalable treatments for migraines, kidney stones, childbirth, cluster headaches, CRPS, and fibromyalgia may be extremely high-impact (cf. Treating Cluster Headaches and Migraines Using N,N-DMT and Other Tryptamines, Using Ibogaine to Create Friendlier Opioids, and Frequency Specific Microcurrent for Kidney-Stone Pain). More research efforts into identifying and quantifying intense suffering currently unaddressed would also be extremely helpful. Finally, if the positive valence scale also has a long-tail, focusing one’s career in developing bliss technologies may pay-off in surprisingly good ways (whereby you may stumble on methods to generate high-valence healing experiences which are orders of magnitude better than you thought were possible).
Thank you!!!
Hey Soeren!
I would disagree for the following reason. For a group to contribute equally it needs to have both its average and its size be such that when you multiply them you get the same value. While it is true that people at the 50% percentile get 1⁄10 of the people at the 90% (and ~1/50 of the 99%), these do not define groups. What we need to look at instead is the cumulative distribution function:
The bottom 50% accounts for 3.17% of incidents
The bottom 90% accounts for 30% of incidents
The bottom 95% accounts for 43% of incidents
What I am getting at is that for a given percentile, the contribution from the group “this percentile and lower” will be a lot smaller than the value at that percentile multiplied by the fraction of the participants below that level. This is because the distribution is very skewed, so for any given percentile the values below it quickly decrease.
Another way of looking at this is by assuming that each percentile has a corresponding value (in the example “number of CHs per year”) proportional to the rarity of that percentile or above. For simplicity, let’s say we have a step function where each time we divide the group by half we get twice the value for those right above the cut-off:
0 to 50% have 1/year
50 to 75% have 2/year
75 to 87.5% have 4/year
and so on...
Here each group contributes equally (size * # of CH is the same for each group). Counter-intuitively, this does not imply that extremes account for a small amount. On the contrary, it implies that the average is infinite (cf. St. Petersburg paradox): even though you will have that for any given percentile, the average below it is always finite (e.g. between 0 and 40% it’s 1/year), the average (and total contribution) above that percentile is always infinite. In this idealized case, it will always be the case that “the bulk is concentrated on a tiny percentile” (and indeed you can make that percentile as small as you want and still get infinitely more above it than below it).
The empirical distribution is not so skewed that we need to worry about infinity. But we do need to worry about the 57% accounted for by the top 5%.
Hi! Thank you for elaborating on what your question is :)
“Bulk” is indeed a very ambiguous term. Would you say 80% is “the bulk”? And 20% is “a small percentage”? If so we would be in agreement. If not, it is more of a wording issue than a matter of substance, I think.
Good catch that the numbers I provided would suggest a power law that just keeps going (e.g. similar to St. Petersburg paradox?). If we use the Cluster Headache dataset, the numbers are:
50% percentile experiences 70 CH/year
80% percentile experiences 365 CH/year
90% percentile experiences 730 CH/year
98% percentile experiences 2190 CH/year
So at least in this case the 90th percentile does get 10X the amount of the 50th percentile. But the 98th and 99th percentile is not as high as 100X, and more like 20 to 50x. So not quite the numbers I used as an example, but also not too far off.
I should also clarify that by frequency I mean the product of ‘how many people have it’, ‘how often’ and ‘for how long’.
Here is the main idea: In Lognormal World, you would see a lognormal distribution for “amount of suffering per person”, “peak suffering per person”, “how long suffering above a certain threshold lasts for each person”, etc.
To illustrate this point, let’s say that each person’s hedonic tone per each second of their life is distributed along a lognormal with an exponent that is a Gaussian with mean x and sd of y. We would then also have that x, across different people, is distributed along a Gaussian with a mean of z and sd of t. Now, if you want to get the global distribution of suffering per second across people, you would need to convolve two Gaussians on the logarithmic pain scale (which represent the exponents of the lognormal distributions). Since convolving two Gaussians gives you another Gaussian, we would then have that the global distribution of suffering per second is also a lognormal distribution! So both at the individual, and the global scale the lognormal long-tails will be present. Now, for you to appreciate the “bulk” of the suffering, you would need to look at the individuals who have the largest means for the normal distribution in the exponent (x in this case). Hence why looking at one’s own individual % of time in extreme pain does not provide a good idea of how much of it there is in the wild across people (especially if one is close to the median; i.e. a pretty happy person).
Adding to what Lucas mentioned (how we are motivated in part by longing/addiction for strong rewards): Suffering and negative reinforcement are correlated but are by no means the same thing. In the case of extreme suffering, there seems to be a point where the pain has already maxed out in terms of negative reinforcement capacity, and anything above it is just senseless suffering. Cluster headaches would not cause any less behavioral suppression if they were 10 or even 100 times less painful. Likewise, our ability to reason about pain and pleasure is limited by our state-dependent ability to imagine it. As I argued in the article, our ability to imagine any pain or pleasure that goes beyond our ability to extrapolate with the qualia accessible to us at the moment is very limited.
The bliss of 5-MeO-DMT or epileptic temporal lobe seizures is as Dostoevsky said “a happiness unthinkable in the normal state and unimaginable for anyone who hasn’t experienced it”. Likewise for extreme pain. So you wouldn’t be able to infer that these states exist (and are much more prevalent than one intuitively believes) merely from observing the patterns of reinforcement from a third-person point of view.
The short answer is—extreme pain is vastly more common than is generally believed. Statistics such as 20% of people in the USA experience chronic pain, with 8% experiencing high-impact chronic pain (interferes with most aspects of life). If indeed we live in Lognormal World, we can expect that the median person will probably have relatively low acquaintance with extreme suffering (until old age), but that the people in the top 10% of sufferers will have 10X the amount, and people in the 99% will have 100X the amount. If we take a person-neutral point of view (i.e. Empty or Open Individualism) and care about “moments of experience” it does not really matter who gets to experience it, at least not morally. There are no diminishing returns per person when it comes to the negative value of suffering (once adaptation has been taken into account).
As with other long-tails, it may seem hard to believe that ”...the bulk of suffering is concentrated in a small percentage of experiences...”. But so it is hard to imagine that there are billionaires out there if all one knows about is the income of one’s family and small circle of friends. Millionaires are rare, but not that rare (about 3% of the population), and we have that in the case of income the bulk of capital is concentrated in a small percent of people (e.g. ~20% of the population controlling ~80% of the wealth, and the top 1% controlling ~45% of it).
Likewise, the research presented here would suggest that in the case of suffering there are “suffering billionaires” out there, and that they account for a much larger % of total suffering than we intuitively would imagine.
Thank you!
The article does focus on the long-tail of intensity and quality of both pleasure and pain rather than frequency. That said, it discusses the Lognormal World as a general principle, which would also predict that the frequency of pain or pleasure would follow a long-tail in addition to their intensity and quality.
This is backed up by the previous article “Cluster Headache Frequency Follows a Long-Tail Distribution”, where we analyzed a survey about Cluster Headache frequency among sufferers, and showed it followed a long-tail (with statistics like “The bottom 80% accounts for 17% of incidents and the bottom 90% accounts for 30% of incidents”, and values ranging from 1 Cluster Headache a year all the way to more than 1,000). We should collect data on e.g. kidney stone, migraine, etc. frequency per individual to see if they also follow a long-tail. Given the general pattern, we suspect they probably do.
It’s great that it works for some people, some of the time. In absolute terms, it is a massive good, so it should be promoted more. Pragmatically it might make sense to emphasize it right now given the low probability that DMT will be approved as a treatment in the next few years, so until then Emgality should be discussed more. That said, yes, in terms of % relief it still is in a completely different class than DMT. That is, it tends to reduce incidence rather than get rid of them, and it is only approved for episodic (rather than chronic) CHs, which account for a relatively small % of the number of CHs experienced, as described in this article (due to the long-tail).
In the article specifically about N,N-DMT as a possible treatment for CHs, Quintin added a rough QALY calculation (I should add that any QALY estimate concerning CHs and other ultra-painful disorders will typically severely underestimate the value of the interventions, given the logarithmic nature of pain scales):
While we believe that traditional metrics such as the QALY do not accurately capture the suffering caused by a cluster headache (see upcoming post on the true pain/pleasure scale), a rough QALY calculation would be as follows (focusing on chronic cluster headache sufferers rather than average, since they compromise up to 83% of total headaches[3]):
Facebook AD campaign:
1. An estimated 370,000 Americans suffer from cluster headaches, 68% of whom are on Facebook[4] (=251,000). About 15% of these suffer from chronic cluster headaches (=37,740). According to Sprout Social, the average estimated cost per click of an ad campaign is $1.72. Assuming 1⁄10 who click are cluster headache sufferers, to reach all chronic sufferers would take (ballpark) $650,000.
2. Assuming about 30% of those who view the ad will pursue the treatment (rough estimate-those who put 2 or less on survey results for questions of legality, difficulty to acquire etc.) and that in 68% of cases it cured or nearly cured their clusters (based on survey results), then the resulting increase in QALYs would be (37,740 people * 0.3 * 0.68) * [0.760 (QALY coefficient) * 1 QALY – ( −0.429 (QALY coefficient)* (0.47QALY)) [5]] = $650,000/7, 404QALY = $87.70 per QALY.
3. These ads could also be targeted to users in countries where psilocybin and DMTare legal for use recreationally, increasing conversion rate. Further targeting could be done on Facebook groups (and other social media groups) which are associated with cluster headache treatment.
I completely agree that the members of a cluster headache subreddit or facebook group are not necessarily representative, and in fact quite likely not representative at all.
I think that the conclusion that the distribution follows a long-tail regardless is still accurate. I reason this based on the following point: even if the probability of participating in the survey increased exponentially as a function of the number of times one experiences CHs per year (or sigmoid at the limit), you would nonetheless not be able to make a Gaussian distribution look like a log-normal. The reason is that the rate at which a Gaussian decreases is proportional to the inverse *squared* of the distance from the mean. So we would still get a net decrease at an exponential rate, which does not produce a long-tail (just a somewhat more bulky tail that still tapers off rather quickly). For it to exhibit a long-tail, the probability of participating in the survey as a function of the number of CHs per year would have to grow *doubly* exponentially, at which point we really run out very quickly of possible participants.
That said, I do agree that there is likely an over-estimation of the frequency, but I would argue due to the above reasons that such over-estimation can’t account for the long-tail.
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.