I am a generalist quantitative researcher. I am open to volunteering and paid work. I welcome suggestions for posts. You can give me feedback here (anonymously or not).
Vasco Grilošø
Hi Michael. Thanks for sharing.
It was Grazianoās papers (2021, 2022, some clarifications in 2020) that made illusionism click for me
I found the article from 2022 clarifying (I have not looked into the other 2), but I am still endorsing (physical) realism. I have found persuasive Hakwan Lauās views about consciousness. So I asked Gemini āWhat would Lau say about strong illusionism?ā. The answer is below. It makes a lot of sense to me. I would be curious to know your thoughts. I wonder whether disagreements between illusionists and realists are largely semantic.
Hakwan Lau is fundamentally opposed to strong illusionism. While he agrees with illusionists that our introspection is imperfect and fails to reveal the physical mechanics of our brains, he completely rejects their conclusion that this makes phenomenal consciousness a āfalse beliefā or a non-existent property.
In his book In Consciousness We Trust, Lau argues that strong illusionism throws the baby out with the bathwater. Here is how he dismantles the strong illusionist position:
1. Macro-Level Summaries are Not āIllusionsā
Strong illusionists (like Graziano, Dennett, and Frankish) argue that because our internal models donāt represent the micro-physical details of our neural wiring, the non-physical āmagicā they report must be a hallucination.
Lau counters this with a functional analogy: macroscopic metrics are real, even if they ignore microscopic details. When a digital thermometer gives you a reading of 20°C, it is not giving you a detailed map of the kinetic energy of billions of individual air molecules. It is giving you a simplified, macroscopic summary. But that doesnāt mean ātemperatureā is an illusion or a false belief.
Similarly, Lauās Perceptual Reality Monitoring (PRM) argues that the prefrontal cortex generates a macroscopic summary of sensory reliability. It evaluates a signal and outputs a tag: This is real and present. Just because that tag lacks the metadata of neurotransmitter release doesnāt make the resulting subjective experience an illusion. It is a highly accurate, biologically real computational output.
2. The Fallacy of the āPerfectā Introspection
Illusionism relies on the premise that for consciousness to be real, it must perfectly match our intuitive, pre-scientific introspective beliefs about it (i.e., that it is magical, ethereal, or non-physical). Because science proves it isnāt magical, illusionists conclude it isnāt real.
Lau points out that this is an unfair standard. Science routinely redefines phenomena without eliminating them. When we discovered that water is HāO rather than an indivisible classical element, we didnāt declare that water was an illusion. Lau argues we should treat consciousness the same way: it isnāt the magical inner theater we intuitively thought it was, but rather a physical, metacognitive reality-monitoring system. Redefining it doesnāt erase its existence.
3. The Evolutionary Reality of āPresenceā
Strong illusionists argue that the āfeelingā of experience is just an evolutionary trick to force an organism to care about its environment.
Lau argues that the brain has a much more specific, computationally vital need for this feeling: it must distinguish reality from its own internal noise. A brain constantly generates internal imageryādreams, memories, imagination, and predictive models. If an organism cannot tell the difference between a memory of a predator and a real predator right in front of it, it will not survive.
For Lau, subjective consciousness is that mechanism. When the PRM system successfully tags a sensory signal as originating from the external world, it generates a feeling of undeniable āpresence.ā That presence is not a trick; it is the correct computational answer to the problem of reality monitoring.
Summary
Lau would say that strong illusionists correctly identify that the brain uses simplified representations, but they make a fatal philosophical error in concluding that simplified representations are false. To Lau, subjective experience is the brainās very real, very physical mechanism for anchoring us in reality.
Hi Cynthia. Thanks for the great context.
I wonder how much the results of the CSES study would change if the management practices were similarly good for both conditions (instead of worse for the cage-free chickens). You replied to my related question below that āMy [your] general sense is that option A leads to a greater welfare increaseā.
Relatedly, I [Vasco] wonder how much welfare varies within production systems. For example, I am interested in knowing which of the following results in a greater increase in welfare. Layers going from:
A. Median furnished cages in the European Union (EU) to median cage-free aviaries in the EU. By median furnished cages in the EU, I mean ones with higher welfare per chicken-year than 50 % of the furnished cages in the EU.
B. 10th percentile furnished cages in the EU to 90th percentile furnished cages in the EU.
Do you have sense of how these compare?
~30% preference studies showing that hens really value the things they can only access in cage-free (nesting boxes, perches, dustbathing, etc)
The above is an argument against barren battery cages, but not against all types of cages? All caged chickens in the European Union (EU) must have āa nestā, ālitter such that pecking and scratching are possibleā, and āappropriate perches of at least 15 cmā. Relatedly, I estimate moving hens from battery to furnished cages increases the welfare of chickens 70.6 % as much as moving hens from battery cages to cage-free aviaries.
Hi Abhi. Thanks for sharing.
Are you considering the risk of research on alternative proteins decreasing the welfare of farmed animals by decreasing the population of farmed animals with positive lives? I estimated cage-free layers and slower growth broilers have negative, but close to neutral lives. I am reasonably confident that caged layers and fast growth broilers have negative lives, but very uncertain about other cases. Here is an example of why this may matter in the context of eggs. 64.3 % (= 1 ā 0.357) of layers in the European Union (EU) were outside cages in 2025. Funding work on alternative proteins does not have an immediate effect. It could be that it mostly reduces the consumption of eggs in 10 years, when the fraction of cage-free layers will likely be higher, and mortality lower due to improvements from experience. So I can see research on plant-based alternatives to eggs leading to fewer layers in the EU with positive lives.
Hi Matthew. Great post. I really like combinations of numbers and story-telling. I think they are a core motivation for people getting involved in effective giving.
Hi Huw. Thanks for sharing.
A vast and rapid injection of surplus labour into the global economy is potentially very bad
It is worth noting the speed and scope of an injection of surplus labour is not independent from how it impacts people. I am sceptical of a vast and rapid injection. However, if it happens, I would still expect it to be good.
In Praise of Job Destruction
Thanks for sharing your thoughts.
Thanks for your suggestion on how to present CEAs, weāll think about this further.
Great.
I think that existing disagregated (non scaled) metrics already allow us to make reasonable guesses for what is cost effective to help animals.
A greater spread of pain intensities would update me (at the margin) towards prioritising very painful welfare issues happening over a short time (in particular, just before slaughter) over less painful ones affecting the whole life of animals
I also wonder whether there are cases where the time in less intense pain is decreased cost-effectively, but the time in more intense pain is increased. In such cases, one would have to rely on views about pain intensities to determine whether there is a reduction in pain. The Welfare Footprint Institute (WFI) estimates that cage-free layers experience less annoying and hurtful pain than ones in furnished cages, but that it is unclear whether they experience more or less disabling and excruciating pain. āThe analysis primarily aimed to estimate the minimum welfare improvement associated with transitioning to cage-free housingā. So it could be that cage-free layers also experience less disabling and excruciating pain. However, if this remains unclear accounting for all welfare issues as accurately as possible, and one believes disabling and excruciating pain are much more intense than annoying and hurtful pain, it could be unclear whether cage-free egg campaigns decrease or increase pain.
I am not confident that we will ever gain lots of clarity on scaling, as there are lots of known unknowns and reasonable disagreements on questions up and downstream from how to scale things (sentience, range, etc.)
I am also not confident. However, I think there is high value of information in making at least one good attempt to quantify the intensity of excruciating pain.
Given the subjective nature of pain, I am not convinced a survey sampling from such a specific group is particularly externally valid (I also donāt know that any survey of humans will ever be externally valid to how weād scale pain across other animals).
I agree comparisons across different pain intensities and species will remain very uncertain.
I replied to the last link here
I replied there.
I think we should aim to have secondary effects of interventions
Great to know. Do you have any concrete plans or timelines? If not, how much funding would you need?
Nitpick. I would not call effects on non-target individuals āsecondaryā. I think they may be much larger than those on the target individuals (in expectation), and āsecundaryā makes it sound like they are less important to consider.
I want to do it in a way that doesnāt unnecessarily penalize/āreward areas where that data is/āis not available
Makes sense.
and to not have the effects of all interventions be dominated by deeply uncertain secondary impacts
What if, given reasonable moral and empirical views, we should in fact be very uncertain about whether practically any intervention is better or worse than nothing accounting for effects on non-target individuals? I think this would be useful to know. The model should not force the conclusion that interventions which have historically been supported by the effective altruism community (like saving human lives, and cage-free egg campaigns) are better than nothing for all reasonable moral and empirical views?
Hi Marcus. Thanks for the reply.
I wonder whether the initial testers included people with views on welfare comparisons across species similar to those that I, @NickLaing, @trammell, and @William_MacAskill have. None of us believe that animals matter exactly 0. However, we also have best guesses that the sentience-adjusted welfare range of shrimps should be much smaller than 0.1 % of that of humans. So our views would be described by an option where animal welfare matters much more than in āOnly humans matterā (where animals matter exactly 0), but still significantly less than in āAnimals matter, but much less than humansā (where the sentience-adjusted welfare range of shrimps is 0.1 % of that of humans).
I think adding an option with sentience-adjusted welfare ranges proportional to the individual number of neurons would let people specify their views. I suspect users would not be much bothered by having 5 options instead of 4. However, if keeping 4 options is important, I would add the option I suggested, and then have a single option replace the current intermediate options (āAnimals matter, but much less than humansā, and āAnimals matter, but somewhat less than humansā).
I understand one could use the advanced mode to specify other ways of comparing welfare across species. However, I am thinking that it is also important for the default options to cover the range of reasonable disagreement, and I believe a large part of that range for welfare comparisons across species is not covered with the current default options.
Hi Ben. Have you considered effects on soil invertebrates? I think eating eggs and chickens may impact them much more or less than chickens.
Hi Morgan. I think you are referring to this.
Scaled SADs involve an extra step, converting all pain estimates into Disabling Pain Equivalents using point estimates of scaling ratios. This is a legacy approach that we keep for information only. In April 2026, we decided to prioritize disaggregated results given our high degree of uncertainty over our pain scaling ratios, and the technical complexity involved in modeling this uncertainty.
Have you considered disaggregating the results even further by not comparing welfare across species? I agree comparisons of pains of different types (annoying, hurtful, disabling, or excruciating) within the same species are very uncertain. However, the same applies to comparisons of pains of the same type across species? I would say comparing 1 h of disabling pain in shrimps with 1 h of disabling pain in humans is much harder than comparing 1 h of disabling pain in humans with 1 h of excruciating pain in shrimps. For an intensity of a given type of pain proportional to āindividual number of neuronā^āexponentā, and āexponentā from 0 to 1, which covers the best guesses than I consider reasonable, 1 h of a given type of pain in shrimps is 10^-12 to 1 times as intense as 1 h of the same type of pain in humans, as shrimps have 10^-6 times as many neurons as humans. Here is some context about my uncertainty.
I would present 6 cost-effectiveness estimates:
Days of annoying pain averted per $ (A).
Days of hurtful pain averted per $ (B).
Days of disabling pain averted per $ (C).
Days of excruciating pain averted per $ (D).
āEquivalent days of disabling pain averted per $ā = āratio between intensity of annoying and disabling painā*A + āratio between the intensity of hurtful and disabling painā*B + C + āratio between the intensity of excruciating and disabling painā*D.
āSADs averted per $ā = āequivalent days of disabling pain averted per $ā*āsentience-adjusted welfare range (as a fraction of that of humans)ā, where āsentience-adjusted welfare rangeā = āprobability of sentienceā*āwelfare range conditional on sentience (as a fraction of that of humans)ā.
People could then change the pain intensity ratios, and sentience-adjusted welfare range to get their own estimates if they want. I believe presenting all the estimates above is useful because people have very different views about not only pain intensities, but also welfare comparisons across species.
At the same time, I would keep the last of the above cost-effectiveness metrics to increase transparency about trade-offs between different pain intensities and species. The trade-offs will still be made even if they are not quantified, and I worry they will be harder to examine and improve on if they are not made explicit.
@Morgan Fairless, @vicky_cox, and @Vince Mak šø, would you find useful a time trade-off (TTO) survey asking people suffering from cluster headaches about the Welfare Footprint Instituteās (WFIās) pain intensities? They may have recently experienced disabling and excruciating pain. I assume the vast majority of people who contributed to Ambitious Impactās (AIMās) estimates of the pain intensities have not recently experienced excruciating pain. So I believe such survey would provide much stronger evidence about the intensity of excruciating pain. I am asking you because AIM and Animal Charity Evaluators (ACE) are the 2 organisations using WFIās pain intensities in cost-effectiveness analyses (CEAs).
All content from our now-closed AIM Research Training Program
I think it is great that you have shared this.
Thanks for the update. You estimate that excruciating pain is 48.0 (= 11.7/ā0.244) times as intense as hurtful pain. This implies 16 h of āawareness of Pain is likely to be present most of the timeā (hurtful pain) is as bad as 20.0 min (= 16ā48.0*60) of āsevere burning in large areas of the body, dismemberment, or extreme tortureā (excruciating pain). In contrast, I think practically everyone would prefer 16 h of hurtful pain over 20 min of excruciating pain.
I agree with all your points. I suspect your guess for the probability of cage-free systems being better than furnished cages is lower than my guess of 2ā3. However, I do not think this matters much. In practice, I would still generally prioritise research on decreasing the uncertainty over cage-free egg campaigns.
As I say in the post, I think that the WFI analysis does not include some potentially significant harms (such as chronic stress/āpain from violence or parasitesāwhich are likely higher in cage-free systems).
It also excludes foot lesions and air quality, which are discussed in section āImportant Considerationā of Chapter 9 of the book Quantifying Pain in Laying Hens by Cynthia Schuck-Paim and Wladimir Alonso from WFI. Here are their conclusions.
[...] the reporting of the relatively low incidence of the more severe and painful manifestations [of foot lesions] in layers [16,21ā23] makes it unlikely that consideration of this harm would affect the estimates substantially to the point of changing any of the conclusions.
[...]
[...] it is not unreasonable to suppose that through the potentially detrimental effect on the respiratory system [24] and on mucous membranes, high concentrations of ammonia can lead to a prolonged state of discomfort [26]. This is an important welfare concern, which is likely to increase the estimated time in pain endured in cage-free facilities, depending on the prevalence of (cage and cage-free) facilities where manure is not regularly removed and the ventilation flow is insufficient, and on the time endured in discomfort (e.g. in temperate regions, higher levels of ammonia in cage-free relative to cage housing have been found in winter, but not in summer [27,28]).
Hi Marcus. Thanks for the clarifications.
I think it would be more productive to make specific critiques of our work and choices because the specific choices are there to see. Itās not costless in time to do this, so I donāt begrudge anyone for not engaging, but you donāt have to infer if you think our animal moral weights are too high, or the cost-effectiveness of an area is too low, you can see what we chose and say what you think is wrong and why.
I very much agree. I would be curious to know your thoughts on these specific critiques.
Hello. Have you considered adding answers to question 1 of the Donor Compass for which animals matter more than exactly 0, but have sufficiently small sentience-adjusted welfare ranges that interventions targeting humans are prioritised? This holds for the answer āOnly humans matterā, but this is not a reasonable view. It requires all animals having a probability of sentience of exactly 0.
The answer where animals matter the least, but more than exactly 0 is āAnimals matter, but much less than humansā. The sentience-adjusted welfare ranges of this answer are below. That of shrimps is 0.1 % that of humans. There are many reasonable values which are above exactly 0, but much smaller than 0.1 %? I would say one could reasonably believe that sentience-adjusted welfare ranges are proportional to āindividiual number of neuronsā, and shrimps have 10^-6 times as many neurons as humans.
I would let users specify sentience-adjusted welfare ranges covering the values given by āindividual number of neuronsā^āexponentā, and āexponentā from 0 to 2. An exponent of 0.188 explains pretty well the welfare ranges in Bob Fischerās book about comparing welfare across species, as illustrated below.
Hi Spencer. Great resource. I set up a reminder to check one view each week, and shared it on EA Lisbonās WhatsApp group.
Hi Jim. The probability of sentience would be relevant if the welfare range dropped very fast as the probability of sentience decreases. For example, if āwelfare rangeā = āprobabiltyā^āexponentā, and āexponentā is very high. For instance, an exponent of 5 would lead to a 10 % chance of sentience resulting in a welfare range of 10^-5, and therefore in a sentience-adjusted welfare range of 10^-6. However, it is unclear to me why the probability of sentience and welfare would related in that way. So the point of your post remains.