--- title: AI for impact needs a place canonical_url: https://ensurance.app/guide/ai-for-impact-needs-a-place markdown_url: https://ensurance.app/guide/ai-for-impact-needs-a-place.md subtitle: a theme is not a unit. a place is category: nature-finance --- # AI for impact needs a place *a theme is not a unit. a place is* AI for impact is a named place, a living outcome on that place, and a limit on what the reading may claim. The outcome needs a way to show whether it moved. A theme collects many efforts under one word, such as climate or biodiversity. The unit is one place, with the outcome and the limit written down. A foundation officer sees the other version in the deck. The model has a name. The benefit is global. The watershed, the forest, or the parcel appears as a photograph, and the person who already keeps it is missing. A river, a forest, a fishery, and the ground a culture already keeps exist whether or not anyone runs a model. A reading can make one of those living things easier for someone who already cares to see. The living thing stays the subject. :::johnson **AI for good is a tool pointed at a living thing someone already keeps.** It is not a promise that the model will save the climate. [the living things are already the subject →](/natural-assets?from=guide) ::: ## what AI for impact is **AI for impact** holds three parts at once. A grant that has the model and the theme, and still lacks the place, has funded a label. The **named place** is specific enough to visit, to map, or to assign to someone already responsible. A basin. A reach of river. A fishery. A parcel. "Nature" and "the climate" can title a program. They leave the ground unnamed. The **living outcome** is what should change, or hold, on that place. Water through the dry months. A fishery that still feeds the people who fish it. A forest watched closely enough that a quiet loss would be seen. The outcome belongs to the living system and to the people already tied to it. The **limit** is one sentence in the file. A reading may sort a map, a species list, a condition, or a dependency. That sentence also sets the ceiling. The model stays a tool. The steward stays the steward. A benefit the place has not shown stays out of the report. Movement is a prior agreement. Before the money, write what would count as the outcome having moved, and what evidence the people on the place would recognize. A condition note the steward already keeps, taken again. A dependency the town already argues about, now on one page. A repeat walk in the season that matters. The people on the place should be able to say whether the note matches what they see. The river does not need the model. The person who loves the river might. ## five tests before the grant Use these before a foundation funds an AI project in the name of impact. A pass is something a person could check. A fail is a theme with a model in front of it. | test | a pass | a fail | | --- | --- | --- | | named place | A watershed, a fishery, or a parcel a person can name | "The climate," "nature," or "global biodiversity" | | what already lives there | The stocks and flows that place already produces, described for that place | A species list or a photograph borrowed from somewhere else | | who already cares | A person, a town, a land trust, or a foundation that kept this place before the model | An audience invented for the slide | | what one reading may claim | It sorted a condition, a map, or a dependency, and it says so | It improved the place, or it is the place | | what it must not claim | The ceiling is written beside the grant | The model saved the climate, offset new compute, or reported a benefit the evidence does not show | The cousin method for choosing the living thing is [pick a natural asset the way you'd pick a GiveWell charity](/guide/pick-a-natural-asset-like-a-givewell-charity?from=guide). This page does not retell it. A forest does not come with a unit borrowed from global health, and a model does not create one. These tests only limit what an AI reading may say about a place you have already been willing to name. A foundation that holds many sites under one theme can still use the table. The honest rollup is a list of rows. Each row carries a place, what lives there, who cares, what the reading claimed, and what it refused. One total for the whole list turns the rows back into a theme. Keep the rows. ## a climate-benefit slide is not a unit Power, water, land, and a local bill are real costs of new compute. [Why communities oppose data centers](/guide/why-communities-oppose-data-centers?from=guide) is that account. On February 17, 2026, [The Guardian](https://www.theguardian.com/technology/2026/feb/17/tech-companies-traditional-ai-generative-climate-breakdown-report) reported a review by energy analyst Ketan Joshi: 154 claims that AI helps the climate. Most describe traditional machine learning, not the generative systems driving new data centers. The public evidence is weak, often a company line rather than a paper. A climate-benefit slide built from that pile fails the tests above. It names no place a person keeps, no living outcome a steward could check, and no ceiling. The trap has a shape. "AI will fix the climate" takes a small machine-learning result and dresses it as a reason to expand generative load. We do not make that claim, including about our own work. The hopeful half is real when it stays small. A model can sort a watershed, a species list, a language archive, or a parcel map. The sort helps while it stays on the named place and with the person who already cares. ## what the reading may say Put this sentence in the grant file, with a real place in it: This funding pays for a reading of this place, kept by these people, aimed at this outcome. The reading may report condition, dependencies, and what would count as movement. It may not report a climate benefit, an offset for compute, or a change the stewards have not measured. If "this place" is still a blank, the file is still a theme. Stocks, flows, and condition are what a reading may list for that place: what is there, what it does, and how it is holding up. The stocks-and-flows map is [natural capital](/natural-capital?from=guide). **RealValue** is our bridge from that inventory to a funder, so a steward and a funder can look at the same place. The worth of the place is the living system. This page invents no dollar amount for it. If the question is the physical place under a training run, read [the model is trained on a basin](/guide/the-model-is-trained-on-a-basin?from=guide). This page does not retell it. The basin under a lab and the place a grant would keep are different subjects. ## the memo after a principal cares The tests are for a foundation that already keeps a place, or funds the people who do. The grant stays as small as that place. Investors enter as the desk that wants a memo after a principal cares. The memo is the row: the named place, the living outcome, the evidence that would show movement, and the sentence the reading must refuse. One place is a complete memo. Several places stay several rows. [Ensurance](/?from=guide) is one worked example of this narrow use. We use the tool to read a place — what it needs, what it depends on, and who depends on it — so a person can fund the care. A [coin](/general?from=guide) is general ensurance, funding across the protocol. A [certificate](/specific?from=guide) is specific ensurance, tied to one natural asset. An [agent](/agents?from=guide) is the account for a place, a people, or a purpose. The instruments are live. Volumes are small. We do not claim a climate-benefit number. The reading is not the river. A foundation that already funds a place, and wants this test kept beside it, can start on the [foundations](/solutions/foundations?from=guide&topic=ai-for-good) page. Bring the place. ## frequently asked questions ### what is AI for impact? AI for impact is a named place, a living outcome on that place, and a limit on what a reading may claim, together with a way to know whether the outcome moved. The model helps someone see the place. The place is the subject. ### how is AI for impact different from a theme? A theme gathers work under a word: climate, biodiversity, oceans. A place can be better or worse this year, and someone is already responsible for it. Funding that cannot name the place is still a theme, including when the budget line says AI. ### what should a foundation ask before it funds an AI project? Ask the five tests, and treat a blank answer as a stop. 1. What is the place, in a name a person can map or visit? 2. What already lives there, and who already keeps it? 3. What living outcome would count as movement, and what evidence would show it? 4. What may this reading claim, in one sentence? 5. What must it refuse, including any claim that the model saved the climate? If the model were gone tomorrow, the place should still be what the grant is about. ## sources [The Guardian, February 17, 2026](https://www.theguardian.com/technology/2026/feb/17/tech-companies-traditional-ai-generative-climate-breakdown-report) — Ajit Niranjan, reporting Ketan Joshi's review. Cited above. ## the series 1. [what AI for good actually is](/guide/what-ai-for-good-actually-is?from=guide) 2. [AI for nature starts with the place](/guide/ai-for-nature-starts-with-the-place?from=guide) 3. [AI for impact needs a place](/guide/ai-for-impact-needs-a-place?from=guide) 4. [AI for the cause you already keep](/guide/ai-for-the-cause-you-already-keep?from=guide) 5. [a tool for the place you already keep](/guide/a-tool-for-the-place-you-already-keep?from=guide)