Welfare Biology and AI: The Quiz

Are you Quiverfull, pro-life, or pro-choice – but for nematodes? A quiz to find your place on the welfare ecology map.

This is part 1 of a five-part sequence on the intersection of welfare biology, land use, and transformative AI. Part 2 covers the empirical landscape. Part 3 covers interventions. Part 4 explores a novel model of invertebrate suffering. Part 5 covers how artificial superintelligence changes everything.

Introduction

There are approximately 4.89 × 10²⁰ soil nematodes on Earth – roughly 57 billion per human. There are about 10¹⁹ soil arthropods (mites, springtails, ants, termites). In the oceans, another 10²¹ nematodes and 10²⁰ arthropods. These are the most abundant animals on the planet by a wide margin, and almost no one thinks about their welfare.

Consider this: If you’re a human (rather than an AI), you had a 1 in 137 billion chance of being born as a human rather than any other possibly sentient being on Earth. The odds of the Sentient Life Lottery are staggering! And with being a human come unparalleled power and responsibility.

Whether we should think about the welfare of all these other beings – and if so, what we should do – depends on a set of ethical and empirical premises that different people hold with very different levels of conviction. Before diving into the science and the interventions, I want to help you figure out where you stand. The rest of the sequence will make much more sense if you know which conclusions follow from your premises, rather than mine.

The Sequence at a Glance

Part 1 (this post) starts with a quiz to help you locate yourself on the key axes of disagreement: population ethics, invertebrate sentience, and what metric to use. Your answers determine which conclusions in the later posts follow from your premises. I then provide a brief orientation to the field of welfare biology and the researchers whose work this sequence builds on.

Part 2: Welfare Biology and AI: Soil and Sea presents the empirical landscape. How many nematodes and arthropods are there, and where? I break down population density by biome (boreal forests have ~7× more soil fauna per m² than cropland), explain why net primary productivity is the master variable, and show why applying pesticides at constant NPP can create a high-throughput killing field rather than a genuinely smaller population. The post also covers marine invertebrates, ocean acidification, and dead zones.

Part 3: Welfare Biology and AI: What We Can Do Now turns to interventions. I unpack Vasco Grilo’s finding that the effects of GiveWell’s top charities on soil animals are ~90,000× larger than their effects on humans, lay out a tiered portfolio of interventions (from funding existing charities to dietary changes to research), and argue that broad-spectrum approaches win over targeted ones because of robustness against trophic backfire and sentience uncertainty.

Part 4: Welfare Biology and AI: The Psychopath, the Nematode, and the Arahant takes a detour into the philosophy of mind. Drawing on M.E. Thomas’s phenomenology of no-self psychopathy and Daniel Ingram’s descriptions of advanced meditative states, I propose that invertebrate pain is probably more like an arahant’s experience of pain than a normal human’s – the nociceptive signal fires, but with no self to amplify it into more comprehensive suffering. This model suggests arthropods (especially social insects with rudimentary self-models) may suffer qualitatively more than nematodes.

Part 5: Welfare Biology and AI: The AI Eats the Sun covers the biggest variable: artificial superintelligence. I work through the Dyson swarm timeline, show that a million replays of 3 million years of evolution at neuron-level resolution takes a Dyson swarm only ~3 years (producing more potential suffering than 30 billion years of Earth’s biosphere), discuss digital suffering in RL agents, and argue that getting the value-loading right may dwarf every other intervention in this sequence combined.

The Quiz

For each question, pick the answer that comes closest to your view. There are no wrong answers – these are genuinely contested questions among thoughtful people. I’ll explain what each answer implies afterward.

Question 1: The Repugnant Conclusion

Imagine two possible worlds:

• World A. 10 billion beings, each living a deeply fulfilling life.
• World B. 100 trillion beings, each living a life that is barely worth living – just slightly more good than bad.

World B contains vastly more total welfare. Do you prefer it?

• (a) Yes. More total welfare is better, even if each individual life is barely positive.
• (b) No, and not even close. I strongly prefer World A – quality matters, not just quantity.
• (c) I don’t want to think about that, but certainly once a mind does exist, it must be protected regardless of its level of happiness.

Question 2: The Asymmetry

Consider two scenarios:

• Scenario X. You can prevent the birth of a being whose life would be full of suffering.
• Scenario Y. You can cause the birth of a being whose life would be full of joy.

Are these morally symmetric?

• (a) Yes. Preventing suffering and creating joy are equally important. (Symmetric.)
• (b) No. Preventing suffering is more important than creating joy. Creating joyful lives is nice but not obligatory. (Asymmetric / antifrustrationist.)
• (c) No. Preventing suffering is important, but creating joyful lives is also important, just somewhat less so. (Weakly asymmetric.)

Question 3: Invertebrate Sentience

A soil nematode (Caenorhabditis elegans) has 302 neurons. It shows nociceptive responses (withdrawal from noxious stimuli), sensitization (increased avoidance after exposure), and classical conditioning (learned association between neutral and noxious stimuli). Rethink Priorities (RP) estimated a 6.8% probability that adult nematodes (or specifically C. elegans) are sentient.

A soil mite has roughly 2,750 neurons. A springtail has roughly 6,000. An ant has roughly 250,000. These all show more complex behavior than nematodes, including social learning in the case of ants.

How do you weigh their potential suffering?

• (a) I think nematodes, mites, springtails, and ants are all probably sentient. The numbers are so vast that we must take their welfare seriously.
• (b) I’m very uncertain. I’d assign something like RP’s probabilities: ~7% for nematodes, ~30% for mites and springtails, ~50%+ for ants. Even low probabilities matter at these population sizes.
• (c) I think ants and maybe springtails are plausibly sentient, but nematodes with 302 neurons are too simple. I’d focus on arthropods.
• (d) I don’t think any invertebrate is sentient in a morally relevant way.

What Your Answers Mean

The Population Ethics Spectrum

Your answers to questions 1 and 2 place you on a spectrum that’s helpful to map onto more familiar ethical territory.

Symmetric total utilitarianism. If you answered (a) to both questions 1 and 2 – you accept the Repugnant Conclusion and see preventing suffering and creating joy as symmetric – then you’re a symmetric total utilitarian. On this view, a world packed with barely-happy nematodes is better than a world with fewer, much-happier elephants. You’d want to increase invertebrate populations if their lives are net positive, and decrease them if net negative. The sign of their welfare is everything.

This is somewhat like the Quiverfull movement in family ethics: more lives are better, so long as they’re positive. The analogy isn’t perfect, but the structural logic – “maximize the number of positive lives” – is similar.

Antifrustrationism and strong asymmetry. If you answered (b) to question 1 and 2 – you reject the Repugnant Conclusion and think preventing suffering is more important than creating joy – then you have something like an antifrustrationist position. On this view, you want to prevent beings from being born into likely-suffering lives, but you don’t have a strong obligation to create happy lives. Even at 50/50 odds of net negative welfare, you’d rather err on the side of preventing births.

This is structurally similar to the pro-choice position in reproductive ethics: The decision-maker (here, us as stewards of ecosystems) should prioritize preventing unwanted suffering over maximizing the number of lives.

Person-affecting or moderate views. If you answered (c) to question 1 and something in between on question 2, you probably have a person-affecting or weakly asymmetric view. You care about existing beings and whether their lives are net positive or net negative, but you recognize that this threshold is hard to pin down and rather err on the conservative side. You don’t feel a strong pull to either maximize populations or minimize aggregate suffering. You might think: “Let’s focus on making existing invertebrate lives better rather than agonizing over population sizes.”

This is the “pro-life” middle ground in our analogy: Don’t kill existing beings, but don’t feel obligated to create new ones either.

The Sentience Threshold

Your answer to question 3 determines which animals you think matter. If you answered (a) or (b), the numbers become overwhelming – there are 50 times as many soil nematodes as soil arthropods, and even at very low probabilities of sentience, the expected moral weight is enormous. If you answered (c), you can focus on the ~10¹⁹ soil arthropods and ignore the nematodes, which simplifies the analysis but still leaves staggering numbers. If you answered (d), you can stop reading here – though I’d invite you to reconsider in part 4, where I draw on the phenomenology of psychopathy to argue that even very simple systems may experience something like pain.

Where I Stand

For transparency, mine are all the (b) answers. I feel closest to antifrustrationism and assign nematodes a low but non-negligible probability of sentience. I don’t urgently want to create new happy nematodes, but I strongly want to prevent future generations of likely-suffering ones from being born.

This means I’m looking for interventions that reduce the population of small, short-lived invertebrates without causing additional suffering in the process. As we’ll see in the coming posts, this is a more constrained optimization problem than it might seem – and AI may change the landscape entirely.

Background: What Is Welfare Biology?

If this whole field is new to you, here’s a brief orientation.

Welfare biology is the study of the welfare of wild animals, introduced as a concept by the economist Yew-Kwang Ng in his 1995 paper “Towards Welfare Biology: Evolutionary Economics of Animal Consciousness and Suffering.” Ng argued that evolutionary dynamics – particularly the r-strategy of producing far more offspring than survive – imply that suffering probably dominates happiness in nature. Most organisms that ever live die young and painfully, and evolution has no reason to make death anything other than agonizing. I sometimes focus on the subfield of welfare ecology, which strikes me as even more neglected.

Brian Tomasik developed these ideas extensively, most notably in his essay “The Importance of Wild-Animal Suffering” for the Center on Long-Term Risk (CLR). Tomasik argued that “the scale of brutality in nature is too vast to ignore” and that the animal advocacy movement needed to expand beyond farmed and laboratory animals.

In a follow-up analysis, “Humanity’s Net Impact on Wild-Animal Suffering,” Tomasik used defaunation studies – the Living Planet Index and Dirzo et al. (2014) – to estimate that the average human prevents ~1.4 × 10⁷ insect-years per year through habitat destruction and environmental impact. This suggests that, on balance, a larger human population reduces wild-animal suffering, and that reductions in wild-insect suffering likely outweigh the suffering of farmed animals even when weighting by neuron count.

However, Tomasik himself flagged a significant tension with these estimates. In his analysis of Krausmann et al. (2013) on human appropriation of net primary production (HANPP), he found that NPPeco – the NPP left over for wildlife – was “almost unchanged between 1910 and 2005” and may even have increased since ~1970 due to CO₂ fertilization. This is “difficult to square with findings that indices of both vertebrate and invertebrate abundance have declined by roughly half since 1970.” Tomasik argues that NPP may be “a more stable measure of wild-animal suffering than an index of animal populations,” because defaunation indices can overstate decline through sampling bias toward declining, larger-bodied species and because they may underrepresent “small and ‘boring’ animals like krill, springtails, and rotifers.”

The upshot: Humanity almost certainly reduces some wild-animal populations (especially vertebrates and flying insects), but whether total animal metabolism – and therefore total suffering – has declined is genuinely uncertain. This tension is important to keep in mind throughout this sequence: The land use effects I discuss in parts 2 and 3 are real at the biome level, but their aggregate global impact may be partially offset by CO₂ fertilization and other factors that increase NPP elsewhere.

Rethink Priorities (RP) has done rigorous work on invertebrate sentience and moral weights, estimating welfare ranges for various species based on neuroscientific evidence. Their moral weight project produced estimates that are widely used in EA cost-effectiveness analyses – including the welfare range of 6.68 × 10⁻⁶ (relative to humans) that Vasco Grilo extrapolated for nematodes. He deferred the caculation to Gemini 2.5, so the derivation is opaque, and the result should not be taken literally.

The Wild Animal Initiative (WAI) conducts and funds research on wild animal welfare, focusing on building the academic field. Especially the work of Simon Eckerström Liedholm has stood out to me as highly relevant.

Most recently, Vasco Grilo’s series of posts on the EA Forum – “Cost-effectiveness accounting for soil nematodes, mites, and springtails” (June 2025), “Animal farming impacts soil nematodes, mites, and springtails hugely more than directly affected animals?” (June 2025), and “More animal farming increases animal welfare if soil animals have negative lives?” (October 2025) – has brought soil fauna into the center of EA cost-effectiveness debates. His key finding: the effects of almost any intervention on soil nematodes, mites, and springtails are orders of magnitude larger than the effects on the intervention’s intended beneficiaries. I’ll unpack his analysis in detail in parts 2 and 3.

And Bentham’s Bulldog, building on Tomasik’s earlier work, has argued that humans in the aggregate likely reduce long-term wild animal suffering by decreasing ecosystem productivity – a position that aligns with much of what this sequence explores, though the HANPP evidence urges caution about the magnitude of this effect.

Next Up

In part 2, “Soil and Sea,” I’ll present the empirical landscape: how many organisms are where, what drives their population sizes, and why the difference between reducing NPP and applying pesticides matters enormously from a welfare perspective.