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 Grilođ¸
Karma: 11,666
I remain sceptical of computational functionalism (CF). It implies some sets of AND, OR, and NOT operations lead to consciousness even if run at the rate of one operation every billion years, which seems very implausible to me.
Time to take AI conÂsciousÂness seriously
Thanks, Jan. Yes, I was looking into an older version of the form.
Here is a post elaborating on why I think expanding agricultural land may increase or decrease the welfare/âsuffering/âhappiness of soil invertebrates.
No worries. Thanks for fixing it. The text looks good now.
Thanks for the relevant nuance, @GV đ¸. The 2026 multiplier would ideally account for the donations of pledges made after 2026 as a result of work in 2026, but this may be difficult to model. It will probably be easier in the future when there is more data on how âtop-of-funnelâ work translates into future pledges.
Hi Ollie. Thanks for the update.
I updated the last sentence of this comment, and some of its text went out of format. I am sharing a print below because I cannot copy the text which is out of format.
Hi Richard. Great post.
Suppose weâre torn between multiple moral theories, accounts of which entities are truly sentient, or other broad âworldviewsâ. How we respond to this uncertainty may be very different depending on whether we opt for maximizing expected choiceworthiness or worldview diversification. The former involves centralized agency, weighing the possible stakes of each option in proportion to the credibility of the theory that assigns it such stakes, and then potentially going âall inâ on whatever option yields the best prospect in expectation. Perhaps a good approach for ideal agents, but (I suggest) too risky for the rest of us. The latter alternative decentralizes and devolves power or resources to an ensemble of subagents representing different philosophical worldviews in rough proportion to their credibility.[3]
Maximising expected choiceworthiness involves diversification because the cost-effectiveness of each intervention decreases with spending. Going âall inâ on the option with the highest marginal cost-effectiveness would only make sense for a very limited budget (until the highest marginal cost-effectiveness decreases to the 2nd highest marginal cost-effectiveness). The optimal number of options supported increases as the spending increases. I wonder if maximising expected choiceworthiness with adequate modelling of how marginal cost-effectiveness decreases with spending would lead to a more principled diversification than worldview diversification as practiced today.
It is also worth keeping in mind that wide uncertainty across interventions and worldviews should push one towards supporting interventions decreasing that uncertainty. I think there is often lots of discussion about how to distribute resources across interventions which involve very few aiming to decrease the uncertainty about the optimal allocation. I would like to see much more research on animal sentience, and, more broadly, on comparing welfare across species.
I donât necessarily want to discourage anyone who is more willing to personally go all-in on a neglected high-impact cause area (shrimp welfare, for example). Especially if you imagine what your ideal moral portfolio would look like at the level of all of society, you might well find that going âall inâ on the most neglected of your ideal cause areas is â on present margins â actually the best way for you to diversify societyâs moral portfolio, and make up for othersâ mistaken neglect of important causes.
The point above may also apply to large funders? Coefficient Giving (CG) directed âover $1 billionâ in 2025, and 1.18 billion $ is 0.001 % of the gross world product (GWP) in 2025 of 118 T$.
Hi Eitan. Thanks for sharing. I am enthusiastic about this initiative.
Its impact is more direct â same animals, better lives. Many animal-welfare interventions can be contested based on their indirect or secondary effects, such as how a diet shift ripples through wild-animal populations, say, or whether a reform just moves production elsewhere. Welfare Tech is comparatively straightforward: it improves the lives of animals already being farmed anyway.
I think interventions improving the lives of farmed animals may increase or decrease the welfare of soil invertebrates much more than they increase the welfare of farmed animals.
I see no escape from the uncertainty about the effects on soil invertebrates if one wants to increase welfare while accounting for all animals. I am not aware of any intervention supported by impact-focussed funders with effects on soil invertebrates robustly smaller than those on the target beneficiaries. Not saving human lives, not decreasing the consumption of animal foods or ingredients, not replacing fast with slower growth chicken, not replacing layers in battery cages with ones in cage-free aviaries or barns, not replacing standard with bird-safe glass, not replacing rodenticide with fertility control bait, and not even electrically stunning farmed shrimps.
Itâs cost-effective, and the economics compound. Take shrimp stunners, estimated to help on the order of 1,400â1,500 shrimp per dollar per year.5 On a welfare-adjusted basis, some analyses (admittedly controversial) found it arguably ~100Ă more cost-effective than cage-free corporate campaigns, which are estimated at 9â120 chicken-years per dollar. Those multiples lean on moral-weight assumptions for shrimp and pain during death; but even discounted, theyâre remarkable.
5. Vasco Grilo, âCost-effectiveness of Shrimp Welfare Projectâs Humane Slaughter Initiative,â EA Forum, 6 Oct 2024: forum.effectivealtruism.org/ââposts/ââEbQysXxofbSqkbAiT; and the 2025 ACE review of SWP. The â100Ă multiple relies on Rethink Prioritiesâ median shrimp welfare range (~3% of a humanâs).
It would be better to argue for high cost-effectiveness comparing interventions targeting similar species (with one intervention involving welfare tech)? Otherwise, the results of the comparisons will be influenced a lot by the uncertainty in welfare comparisons across species. I have practically no idea about whether the Shrimp Welfare Projectâs (SWPâs) Humane Slaughter Initiative (HSI) has increased the welfare of shrimps more or less cost-effectively than cage-free egg corporate campaigns increase the welfare of chickens. I estimated HSI has increased the welfare of shrimps 139 times as cost-effectively as cage-free egg corporate campaigns increase the welfare of chickens. However, for sentience-adjusted welfare ranges proportional to âindividual number of neuronsâ^âexponentâ, and âexponentâ from 0 to 2, which covers the best guesses that I consider reasonable, I calculate that HSI has increased the welfare of shrimps 2.26*10^-4 to 1.49 k times as cost-effectively as cage-free egg corporate campaigns. interventions increases the welfare of their target beneficiries more cost-effectively. Below is a graph with the results for âexponentâ from 0 to 2.
Hi Aidan. Thanks for the update. This looks like a great analysis, as did the ones for 2020-2022, and 2023-2024.
We note that our multiplier will face pressure from the denominator in coming years. Our 2026 budget of
The increased spending in 2026 will be useful to get a better sense of the marginal multiplier. The multiplier on the total spending over 2026 being significantly lower than 7 (the multiplier on the total spending over 2025) would signal a marginal multiplier in 2026 significantly lower than 7.
The misÂtake of conÂflatÂing inÂtelÂliÂgence and power
Can you provide some specific examples, and Iâll reconsider
I mostly had this section in mind. I think the positive correlation between hedonic welfare and reported seemingly honest preferences in humans provides some weak evidence that there is such a correlation for AIs. There are some similarities between humans and AIs.
Iâm not sure if we âagreeâ here, or at least thatâs not the point I was making. I was accepting that AI models (either the information flow itself, or the physical electron flows or something) might indeed âhaveâ or generate consciousness and welfare.
Right. However, I think AI welfare being difficult to measure conditional on sentience tends to imply that assessing AI sentience is also difficult. So I would prioritise decreasing uncertainty about this too.
What I meant was that even if the AI is making a confident statement that it is in pain, and it is not deliberately lying, it doesnât mean that it actually knows the answer to âis it (or is anything in itâs system) in painâ.
So the âeven ifâ was setting off a contrast between ânot having a access to the answerâ and âmaking a confident statementâ.
Makes sense.
In addition, in allometry, âthe study of the relationship of body size to shape,[1] anatomy, physiology and behaviourâ, âThe relationship between the two measured quantities is often expressed as a power law equation (allometric equation)â.
From the book âScale: The Universal Laws of Growth, Innovation, Sustainability, and the Pace of Life in Organisms, Cities, Economies, and Companiesâ by Geoffrey West:
This scaling law for metabolic rate [this one], known as Kleiberâs law after the biologist who first articulated it, is valid across almost all taxonomic groups, including mammals, birds, fish, crustacea, bacteria, plants, and cells [see this section of my linkpost for further discussion]. Even more impressive, however, is that similar scaling laws hold for essentially all physiological quantities and life-history events, including growth rate, heart rate, evolutionary rate, genome length, mitochondrial density, gray matter in the brain, life span, the height of trees and even the number of their leaves. Furthermore, when plotted logarithmically this dizzying array of scaling laws all look like Figure 1 and therefore have the same mathematical structure. They are all âpower lawsâ and are typically governed by an exponent (the slope of the graph), which is a simple multiple of Âź, the classic example being the ž for metabolic rate. So, for example, if the size of a mammal is doubled, its heart rate decreases by about 25 percent. The number 4 therefore plays a fundamental and almost magically universal role in all of life.13
Footnote 13:
There are several excellent texts summarizing the various allometric scaling laws in biology. Among them are: W. A. Calder, Size, Function and Life History (Cambridge, MA: Harvard University Press, 1984); E. L. Charnov, Life History Invariants (Oxford, UK: Oxford University Press, 1993); T. A. McMahon and J. T. Bonner, On Size and Life (New York: Scientific American Library, 1983); R. H. Peters, The Ecological Implications of Body Size (Cambridge, UK: Cambridge University Press, 1986); K. Schmidt-Nielsen, Why Is Animal Size So Important? (Cambridge, UK: Cambridge University Press, 1984).
Hello. Thanks for sharing. Yesterday I created a WhatsApp group for people interested in effective altruism in Porto. It is under the WhatsApp community Effective Altruism and Rationality Portugal, where I just shared the post above.
You may want to add deadlines for submitting the application forms, and consider having a single form for all applications (where people can specify the location they are interested in).
ExÂpandÂing agriÂculÂtural land may inÂcrease or deÂcrease the welfare/âsufferÂing/âhapÂpiness of soil invertebrates
I am very uncertain about whether decreasing the number of farmed animals increases or decreases animal welfare (in expectation) due to potentially dominant effects on soil invertebrates. In addition, I worry decreasing the number of farmed animals may prevent some from having positive lives.
Hi David. Great post. I broadly agree. Relatedly, readers may be interested in the post AI Welfare Is (Frankfurtian) Bullshit.
I sometimes felt you were implying âno evidenceâ when I thought something closer to âvery weak evidenceâ would be more appropriate. However, in practice, I would agree the focus should overwhelmingly be on decreasing the uncertainty about the extent to which AI models have welfare, and how to measure it.
I donât find this convincing. (NB: Iâm still working on this response.) Even if the talker has no access to the valenced consciousness, itâs model may simply lead it to a confident and wrong answer about this.
Did you mean âIfâ instead of âEven ifâ?
Wha tmight
Nitpick. âWhat mightâ.
Hi AĂŻda. That makes sense. However, I think the reasons you mentioned sometimes play against increasing animal welfare. Poorer welfare standards tend to have a lower carbon and land footprint.
Hi Ollie. I can no longer remove tags in my draft or published posts.