Effective altruism is the project of using evidence and careful reasoning to work out how to help others as much as possible with what you have, and then actually doing it. If you arrived through GiveWell, Giving What We Can, or an employer match that pointed you at a donor-advised fund, you already run the operating system: evidence over anecdote, expected value over warm glow, cause neutrality over whichever charity found you first.
That operating system is one of the best things to happen to giving in a century. This post does not argue with it. It points at an input the OS has never priced, because the input never showed up as a cause.
the operating system, in its own words
Strip away the forum, the community, and the discourse, and effective altruism rests on three commitments.
Evidence. Prefer what is measured to what is felt. GiveWell publishes its cost-effectiveness models down to the cell and revises them in public when the numbers move. The randomized trial became a fundraising document because this movement asked it to.
Expected value. Weigh outcomes by probability times magnitude, and stay scope-sensitive: a thousand lives is not one life with a bigger photo. This is what lets the same donor fund a malaria net today and a small chance of averting a very large catastrophe tomorrow without feeling like two different people.
Cause neutrality. Give where the marginal dollar does the most, not where the story is best. Importance, tractability, neglectedness. Be willing to move.
Run honestly, these produce a portfolio, and the portfolio is real. Global health and wellbeing: malaria nets and seasonal chemoprevention, vitamin A, unconditional cash. Animal welfare: corporate cage-free campaigns that changed how the largest food companies buy eggs, and a growing research program on wild animals. AI safety and existential risk: alignment research, biosecurity, and the institutions meant to catch a civilization-scale mistake before it lands. A child who does not get malaria is real. A hen out of a battery cage is real. A lab that takes alignment seriously because someone funded the work is real. None of that is what this post is about.
One governance sentence and we move on: the FTX collapse in 2022 was ends justifying means with nobody auditing the means, the movement's own postmortems said so, and the lesson is about guardrails, not about you.
where the living world got filed
Ask the operating system about nature and you get two answers. Both of them lose it.
The first answer is climate. Here EA holds a defensible position: climate is important but not neglected. 80,000 Hours ranks it below its top problems on exactly that ground, since hundreds of billions of dollars a year already flow toward it. Fair. But "climate" is one flow out of a living system, the carbon one. Money into panels and policy does not keep a watershed making water, a grassland making forage, or a soil holding the rain when it finally comes. Filing nature under climate is filing the whole factory under one product.
The second answer is charity. Conservation reads as sentiment: a species on a poster, an acre with a plaque, the identifiable-victim reflex the movement trained itself out of. Rigor drops, and the allocation drops with it.
There is a third thing the living world actually is, and it is not on the list.
| how nature gets filed | what the framework hears | what it misses |
|---|---|---|
| climate | crowded, not neglected | carbon is one flow; water, forage, soil, and species are others, and those are not crowded |
| charity | sentiment, weak evidence | the dependency is physical, not emotional |
| infrastructure | nothing yet; there is no line for it | every other cause on the list runs on it |
If you want the neglectedness test in numbers: the UN Environment Programme counted about $220 billion flowing into nature-based solutions in 2023, against $571 billion a year needed by 2030, while $7.3 trillion flowed the other way into nature-negative activity (UNEP, State of Finance for Nature 2026). Private nature-based-solutions finance was about $23 billion — roughly a 30-to-1 tilt the other way. This is what a neglected cause looks like when it is hiding inside a crowded label.
the factory
The subtitle is literal. A bed net is a manufactured object, and the child sleeping under it eats grain grown in soil that rain fed. A set of model weights is a manufactured object, and it was trained in a hall cooled by water drawn from a basin, on a grid that leans on rivers and reservoirs for its own cooling and storage. Both are products. Both assume the plant is running.
Effective altruism is very good at output per dollar, product line by product line. It has rarely funded the plant as the plant, because the plant does not appear as a product line. Climate and land-use grants exist; they still usually file under climate, not as the factory every other cause sits on. It appears as a coefficient on every line at once. Coefficient Giving, formerly Open Philanthropy, chose its name because a coefficient multiplies whatever it is paired with. The living world is the coefficient on the whole sum. Let it slide and every term in the expected-value calculation shrinks together, including the ones you were proudest of.
That is why this is not a pitch for a seventh cause area. Watersheds, forests, soils, species, and the water cycle exist whether or not anyone funds them; they are the substrate every cause on the list assumes for free. Treating the substrate as optional is a choice. It is not neutrality.
You might be thinking: nature is not an existential risk. Toby Ord's The Precipice puts the chance of an existential catastrophe from climate change this century at roughly one in a thousand, and about the same for other environmental damage. Take those numbers at face value. The argument here does not run through extinction. It runs through dependency. You do not need a civilization-ending scenario to notice that the cooling water for the lab that just made you liquid comes from somewhere specific, and that somewhere is not in anyone's portfolio.
You might also be thinking: conservation cannot be measured like a bed net. Partly true, and less true every year. Condition is measurable now: streamflow, soil moisture, canopy, forage, species counts, from satellites, sensors, and boots. And the bar was never "randomized trial or nothing." AI safety has no control group. The bar was best available evidence and honest uncertainty, and a named place with a measured condition is honest uncertainty, which is already more than a lot of giving admits.
And yes, we are pitching something. The thing we are pitching is a place, not a fund. Name a place, not a theme. A live example we already write is inland wetlands — a named living stock, not a cause area. Another is the Okefenokee.
what cause neutrality would actually do
Cause neutrality was a promise to follow the evidence into causes that felt unglamorous. Applied to the substrate, it asks for three moves.
Reclassify. Move the living world out of "climate" and "charity" and into infrastructure, where water utilities, insurers, and farmers already keep it. That is the mental model the next post in this series builds out.
Pick a unit. Not a theme. A named place with a named living stock: a basin, a forage system, a soil, a stretch of river, with a condition you can measure and an additionality question you can ask. A later post in this series walks a GiveWell-style pick, test by test.
Hold it. Do not park it. A donor-advised fund is a parking lot. A living system is a place. The deduction posts the day the shares move; the place is only safer the day something is held on it.
| a cause area | a named living stock | |
|---|---|---|
| what you fund | a theory of change | a place with a condition |
| the unit | an intervention: a net, a transfer, a paper | a stock and its flows: acre-feet, forage, soil, species |
| how you know | a trial or a model | condition over time, measured in the open |
| time horizon | the grant cycle | as long as the place |
| who else pays | other donors | the people downstream: cities, utilities, farms, data centers |
This is where ensurance comes in, and only here. Ensurance is the mechanism that lets a grant or a hold sit on a named natural asset, with its condition, its routing, and a receipt that keeps updating, all in the open. It funds the living thing. It is not the living thing. The people who steward the place govern it; ensurance makes the money legible to them and the place legible to you.
We are early, and we say so. Taking a grant straight from a DAF sponsor is a design target for us, not a shipped button; today that gift usually routes through a land trust or fiscal-sponsor cousin, and we will tell you which one. What you can do now is look at the named places already on the map and start the conversation. None of this is tax, legal, or investment advice; the mechanics of a stock gift belong to your advisor and to the post in this series that covers them.
frequently asked questions
what is effective altruism?
Effective altruism is a research field and a community that uses evidence and careful reasoning to find the most effective ways to help others, then acts on them. In practice it means comparing causes and charities by expected impact per dollar rather than by sentiment, and being willing to move money when the evidence moves. Its best-known institutions are GiveWell, Giving What We Can, 80,000 Hours, and Coefficient Giving (formerly Open Philanthropy).
why doesn't ea fund nature?
Mostly because of how nature is filed. Counted as "climate," it fails the neglectedness test: hundreds of billions already flow to carbon. Counted as "charity," it fails the rigor test: conservation looks like sentiment. Counted as what it is, physical infrastructure that every other cause depends on, it passes both. Nature-based solutions received about $220 billion in 2023 against $571 billion needed by 2030, and only about $23 billion of that was private (UNEP 2026). The living world is neglected; it has just been neglected under a crowded label.
is conservation cost-effective?
It can be, and it can be measured. Protecting intact ecosystems is among the cheapest moves in the UN's nature-finance accounting, but cheapness per acre is not the same as outcomes per dollar — the question GiveWell actually asks. US land trusts helped conserve about 61 million acres (2020 census): roughly 20 million under easement and 8.5 million owned, plus acres reconveyed to agencies. That is a real unit: a named parcel, a recorded restriction, annual monitoring. What a distant donor often still lacks is a receipt that keeps showing the place. That is the narrower gap this series is about, and no one here is claiming a dollars-per-life number that beats a malaria net. The claim is smaller and harder to dodge: the net assumes the world that makes the water.
read next
- nature is infrastructure. a DAF can fund it — the reclassification, worked out
- explore named places already on the map
- how to fund a conservation group — if you already have a land trust in mind
the series
Twelve posts for people about to have more money than plan, most of it headed for a donor-advised fund. Read in order or jump.
- the gift isn't finished when the deduction posts
- a donor-advised fund is a parking lot
- give the shares, not what's left after the tax
- effective altruism forgot the factory — you are here
- nature is infrastructure. a DAF can fund it
- how to grant from a DAF to a place
- longtermism still needs a living present
- pick a natural asset the way you'd pick a GiveWell charity
- the warehouse is the product unless you choose otherwise
- the model is trained on a basin
- some of this should be a grant. some of it should be a hold
- what you actually hold after the grant