Core thesis
Responsible AI cannot stop at protecting people from the risks a model creates. It has to account for what AI takes from — and returns to — the Earth system that makes it possible at all.
Adoption comes first. When adoption scales, it hardens into protocol. And when protocol matures, it has to be able to measure something specific: what AI gives back to the planet it depends on.
That mandate already exists in outline. UNESCO's Recommendation on the Ethics of Artificial Intelligence places environment and ecosystems inside the same policy frame as human dignity, oversight, transparency and fairness. Microsoft's 2026 Environmental Sustainability Report is blunter still: AI is driving new demand for energy, water, land and materials, and innovation at this scale has to be matched by responsibility at the same scale. Independent research on the environmental cost of AI goes further — AI is not only a digital technology. It is a material one, with carbon, water, land, infrastructure and supply-chain costs that can be measured.
[UNESCO — Recommendation on the Ethics of Artificial Intelligence] · [Microsoft — 2026 Environmental Sustainability Report] · [Environmental Cost of Artificial Intelligence — carbon, water & land footprints]What is the Earth Return Index
AI grows.
Earth gains.
ERI — the Earth Return Index — measures how much value AI returns to Earth relative to the resources it consumes.
Put more formally: it evaluates whether AI's growth is matched by verified contributions — to clean energy, carbon reduction and removal, water replenishment, material circularity, ecosystems, the communities that host its infrastructure, and the generations who inherit all of it.
The seven returns
Seven pillars. Seven open questions. Each one asks the same thing in a different currency.
Clean Energy Return
Does AI growth accelerate clean or carbon-free energy?
EvidenceNew capacity, PPAs, grid investment, 24/7 CFE matchingCarbon Return
Does AI help reduce or remove emissions?
EvidenceRemoval contracts, avoided emissions, carbon-negative progressWater Return
Does AI replenish or reduce water impact?
EvidenceReplenishment, reuse, watershed restoration, WUE gainsMaterial Return
Does AI reduce material pressure?
EvidenceCircular materials, low-carbon materials, e-waste reductionEcosystem Return
Does AI support nature and biodiversity?
EvidenceLand protection, biodiversity monitoring, nature-based solutionsCommunity Return
Does AI infrastructure benefit host communities?
EvidenceLocal benefit, energy affordability, jobs, educationFuture Generation Return
Does AI build capability for young people and the public good?
EvidenceYouth participation, NGO involvement, open climate data & toolsWhy the index is missing
Most Responsible AI frameworks today are built around Model Safety, Security, Bias & Fairness, Privacy, and Compliance. Necessary — and not enough. They govern the behaviour of the model. They rarely reach the systems that make the model possible: energy, water, materials, land, communities, ecosystems, future generations.
NIST's AI Risk Management Framework manages risk to individuals, organisations and society across design, development, use and evaluation. OECD.AI tracks policy, accountability, and incidents at global scale, with AI's compute and environmental footprint now entering the frame. Both are rigorous. Neither yet asks the question this dossier keeps returning to:
"What does responsible AI owe back to the Earth that enables it?"
Microsoft Research's proposed Earth alignment principle is the closest existing answer so far — a call to steer AI development and deployment toward planetary stability, alongside its parallel work on AI as both a tool for, and a cost to, sustainability.
[NIST — AI Risk Management Framework] · [OECD.AI Policy Observatory] · [Earth Alignment Principle for AI — Microsoft Research] · [Sustainability Science and Innovation — Microsoft Research]Earth Return Ratio
Earth Value Returned
Clean energy, carbon removal, water replenishment, material circularity, ecosystem restoration, community benefit.
AI Resource Burden
Electricity, emissions, water, land, materials, community pressure.
For public engagement, the working proposal is an ERI Score, 0–100 — legible now, precise later. Or, for the youth audience this research is built for, said in one line:
AI grows.
Earth gets paid back.
Where global protocol already stands
The field is moving from principle → risk framework → auditable management system → continuous monitoring. Here is where the load-bearing pieces already sit.
Five core insights
Responsible AI is moving from ethics to operating system.
"Protocol" is the missing bridge between adoption and responsibility.
The best scorecard is not a single score.
Global monitoring must combine policy, incidents, and evaluation.
The sustainability lens is still the underdeveloped one in AI risk scorecards.
For the roundtable
YUCROWN C// — AI For Youth Research
Nobody is currently framing Responsible AI this way. That is Earth LinC's opening — and the questions below are where we'd start the room.
If every AI lab had to publish one ERI pillar publicly this year, which would move fastest — and which would expose the most?
Can a 0–100 ERI Score ever be more than a marketing number, or does it need the harder Earth Return Ratio underneath it from day one?
Whose verification do we trust — auditors, regulators, open data, or the communities living next to the infrastructure?
What would it take for "future generation return" to be measured as rigorously as carbon or water?
Should Earth Return obligations scale with model size and compute, the way carbon accounting scales with emissions?
What is the smallest disclosure any AI organization could make this year, without waiting for a global standard to force it?