A University of Arizona · BIO5 Institute
PH&AI Summer School · Day 3 · Open Mic
Closing Session · The Footprint & The Rules

What does AI
cost the Earth?

A frontier model has a body — silicon, electricity, water, copper, land. We start with an honest accounting of that cost, then ask the question it forces: who writes the rules, and who pays the bill?

Hosted by Tyson Swetnam, PhD
Associate Professor · BIO5 Institute
Grand Challenges Research Building
Tucson, AZ · June 2026
APH&AI · AI & the Earth
02 / Format
How This Session Runs

Sixty minutes. Open mic. Your questions.

10′
Framing.

Tyson sets the question, the vocabulary, the stakes. The slides that follow are the shared substrate — not a lecture.

40′
Open mic.

The floor is yours. Ask anything about the ethics and the law of AI today — the seed prompts are only a starting point, and disagreement with them is fair game.

10′
So what?

We tie it back to your own work — what actually changes in your own practice. Bring the harder question, not the safe one.

On the table — ask about any of it
Ethics of AI Rights & the law Regulation & liability Corporate vs public rule-making Environmental cost Public-health practice
APH&AI · AI & the Earth
03 / The Bill
Start With The Physical

Every prompt has a body — and somebody pays for it.

The bill

Silicon, electricity, water, copper, land.

A frontier model is not weightless. Training and inference draw megawatts; cooling evaporates freshwater; the chips and the buildings carry carbon paid before the first token. Part 01 itemizes it.

Not just personal virtue

"Ethics of AI" ≠ "Ethical AI."

Choosing a smaller model is an ethical-AI behavior question. Who is accountable for the grid, the water table, and the siting is an ethics-of-AI governance question. The footprint forces the second — Siau & Wang (2020).

Two halves, one conversation: the bill nobody itemized — then the rules nobody agreed on.

APH&AI · AI & the Earth
Part 01 · Footprint
Part 01

The bill
nobody
itemized.

A frontier model has a body. Silicon, electricity, water, copper, land. What an honest accounting of the environmental cost looks like — and where it lands.

APH&AI · AI & the Earth
05 / Where The Cost Lands
Four Inputs · One System

Every prompt sits on a stack of physical inputs. Worth naming them out loud.

Electricity.

Training a frontier model and serving its inference both draw megawatts continuously. Datacenter demand is now a material driver of grid planning in several U.S. regions.

💧
Water.

Evaporative cooling consumes potable or industrial water — sited disproportionately in arid regions. The Southwest is a relevant case in this room.

Embodied carbon.

Accelerator manufacturing, server build, building shell, fiber, copper. The carbon cost is paid before the first prompt — and is rarely on the accounting.

🏗
Siting & land.

Hyperscale datacenter buildout reshapes local economies, tax bases, transmission corridors, and noise envelopes. The externalities are not evenly distributed.

Frame: AI's environmental cost is not one number. It is at least four numbers, on at least three timescales (training, inference, buildout).

APH&AI · AI & the Earth
06 / Electricity
Input 01 · Electricity & The Grid

Data-center demand is doubling — and AI is the accelerant.

Global data-center electricity

~415 → 945 TWh by 2030

2024
2030

IEA's Base Case: data-center use roughly doubles to ~945 TWh by 2030 — just under 3% of global electricity — growing ~15%/yr, four times faster than demand overall. AI "accelerated servers" grow ~30%/yr. — IEA, Energy and AI (2025).

Per query · one text prompt

~2.9 Wh vs ~0.3 Wh

Web search
AI query

An early estimate put a chatbot query near 2.9 Wh — roughly 10× a web search (EPRI / IEA, 2024). Newer medians run lower (~0.3 Wh), long reasoning and multimodal far higher. The exact figure is contested; the direction is not. (Bars illustrative.)

APH&AI · AI & the Earth
07 / Water
Input 02 · Water & Arid-Region Siting

Cooling evaporates freshwater — and the Southwest is in the room.

The thirst of one model

It drinks while it trains.

≈ 70× GPT-3

GPT-3's ~700,000 L (2020) is the last training figure anyone published. GPT-4 is estimated near 70× that — tens of millions of liters, much of it a single month's cooling draw at one Iowa cluster.

undisclosed

For today's frontier models — GPT-5-class, Claude Opus 4.x — the training-water cost is not published at all. The footprint grows; the transparency doesn't.

— GPT-3: Li et al. (2023); GPT-4: estimates from reporting on Microsoft's Iowa training cluster (2023).

Where it lands · Arizona

Sited where water is scarcest.

Goodyear
Phoenix area
Google '24

Microsoft's Goodyear campus drew ~56M gal/yr of potable water (~670 homes) and is now switching to zero-water air cooling amid local limits; Phoenix-area direct cooling runs ~385M gal/yr; the most water-intensive Google site hit ~1 billion gal in 2024. — APM Research Lab; Microsoft; Google (2024). (Bars illustrative.)

APH&AI · AI & the Earth
08 / Carbon
Input 03 · Carbon — Operational & Embodied

Some carbon is burned to run the model. More is spent before it runs.

Operational · one training run

~552 tCO₂e to train GPT-3

1,287 MWh

Electricity to train GPT-3 (2020) — about 552 tonnes CO₂e, roughly the annual tailpipe emissions of ~120 U.S. cars. Frontier models since are substantially larger.

— Patterson et al. (2021). Car comparison illustrative.

Embodied · paid before the first prompt

Chips, servers, shells, copper.

Operational
Embodied

Accelerator fabrication, server build, the building shell, fiber and copper carry carbon spent up-front. For modern hardware it increasingly rivals operational emissions — yet routinely never reaches the ledger. — Gupta et al., "Chasing Carbon" (2021). Split illustrative.

APH&AI · AI & the Earth
09 / Marginal Cost
A Per-Call Heuristic

Two questions worth asking before every agentic loop.

Q1 · Right-size the model

Does this task actually need the frontier?

Tiny / local
Small hosted
Mid
Frontier

Per-call energy and embodied compute scale with model size. Most tasks fit inside a smaller model — defaulting to the frontier is a habit, not a requirement. (Figure is illustrative, not measured.)

Q2 · Right-size the loop

Does this agent need to retry forty times?

One-shot
+ retries
+ tool calls
Speculative agent

Agentic loops fire speculative calls, retries, and tool calls behind one user action. Cumulative compute, not the visible turn count, is the cost. Cache. Cap retries. Watch the loop.

APH&AI · AI & the Earth
10 / Bridge
Bridge · The Two Halves Are The Same Conversation

Environmental impact is an ethics question. Same publics. Same diffuse harms.

Collective, diffuse, opaque harms that older rights frameworks address only partially.

Nelson · on algorithmic harms — and equally on environmental ones
Same structural pattern

Datacenter siting, grid load, and water draw produce harms that are collective, diffuse, and opaque — the same shape as algorithmic harm. Marshall's social-rights tier is the same conceptual address.

Same convergence

Laudato Si' (2015, on care for creation) and Magnifica Humanitas (2026, on AI) share a single underlying claim: the environment and the economy and the technology are one moral question.

APH&AI · AI & the Earth
Part 02 · Who's Accountable
Part 02

Who writes
the rules?

Corporate AI constitutions and public AI bills of rights are doing very different work. Alondra Nelson argues one of them is structurally insufficient.

APH&AI · AI & the Earth
12 / Two Documents
Two Foundational Documents · Two Sources Of Legitimacy

A corporate constitution and a public bill of rights are not the same kind of object.

Type 01 · Corporate AI Constitution

Internal training and alignment spec.

The model developer's vision of how its own model should behave. Canonical example: a frontier-lab "constitution" document used in training.

  • Written and revised by company fiat
  • Not negotiated with affected publics
  • Newer revisions can quietly remove references to international human-rights frameworks
  • Legitimacy comes from the company's own authority
Type 02 · Public AI Bill Of Rights

Democratic rights claim against algorithmic power.

Developed through public process. Canonical example: the White House OSTP Blueprint for an AI Bill of Rights, October 2022.

  • Declares principles publics can extend to new institutions and new harms
  • Five principles: safety, discrimination protections, data privacy, notice/explanation, human alternatives
  • Force comes from democratic legitimacy and the surrounding legal infrastructure
  • Structurally closer to a Declaration than to a Bill — declares, but does not enforce

Source: Alondra Nelson, A civic grammar for AI rights, Science (2026). DOI 10.1126/science.aeh7153

APH&AI · AI & the Earth
13 / Civic Grammar
Nelson's Argument · Civic Grammar Meets Marshall

A shared "civic grammar" of rights — landing on Marshall's social tier.

Non-discrimination, transparency, data privacy, notice, human alternatives — a civic vocabulary traveling across partisan lines by institutional diffusion (Strang & Meyer), mapping onto Marshall's third wave of citizenship rights:

18th c.
Civil rightsWave 01
Liberty of person, speech, faith, property. Individual freedoms against arbitrary state power.
19th c.
Political rightsWave 02
The franchise. Participation in collective political authority. Universal suffrage as the long-running expansion.
20th c.
Social rightsWave 03
Economic security and the conditions of participation. Entitlements against harms of industrial capitalism that civil-liberties frameworks could not address.
21st c. ?
AI rightsNelson's argument
Entitlements against systems "that increasingly govern access to employment, credit, healthcare, housing, and education." Collective, diffuse, opaque harms — a social-citizenship response to algorithmic and environmental power.
APH&AI · AI & the Earth
14 / Limits
What Civic Grammar Cannot Do On Its Own

Three failure modes Nelson is clear-eyed about.

01
Accommodation can mimic transformation.
A vocabulary that moves easily across partisan lines may have been "drained of the political content that gives rights claims their force." Crossing the aisle is not the same as biting.
02
Rights individualize structural problems.
Frameworks built around individual claims often fail to address the collective and systemic nature of algorithmic harms. A model isn't biased against you — it's biased against a class you sit inside.
03
Declaration is not delivery.
"Declarations of entitlement and their substantive delivery can remain decades apart, separated by the organized power of those who benefit from the status quo."
Nelson's question
Will democratic institutions take this civic grammar seriously before the AI companies finish writing their own constitutions for us all?
APH&AI · AI & the Earth
15 / Encyclical
A Moral Response · Pope Leo XIV · First Encyclical (2026)

Magnifica Humanitas: Nelson's verdict, reached through a very different door.

Encyclical · Lat. Magnificent Humanity

Magnifica
Humanitas

Signed 15 May 2026
135th anniv. of Rerum Novarum (1891)
Addressed "every person of goodwill"
The teaching

Human dignity affirmed as infinite. Transhumanism (engineering away finitude) and posthumanism (blurring the human/machine line) are named an "anti-human vision" — measured against dignity, common good, justice.

The demands

Disarm AI from military and purely economic ends; subject companies to stricter state and international regulation — explicitly not industry self-governance.

Convergence
Leo XIV and Nelson reach one conclusion: corporate self-governance is structurally insufficient as a source of legitimacy.
APH&AI · AI & the Earth
16 / Declarations
A Working Map · Not Exhaustive

Major declarations and agreements you can name from memory.

Document Issuing body Force What it does
Framework Convention on AI & Human Rights Council of Europe Treaty Binding international convention tying AI systems to human-rights, democracy, and rule-of-law obligations.
Recommendation on the Ethics of AI UNESCO Recommendation First global standard-setting instrument on AI ethics, adopted by 193 member states.
OECD AI Principles · G20 endorsement OECD · endorsed by G20 Principles Trustworthy-AI principles widely adopted as a starting point for national strategies; signal of cross-bloc consensus.
Political Declaration on Responsible Military Use of AI & Autonomy U.S. State Dept. + co-signatories Declaration Norms for military AI: human accountability, legal review, controlled escalation.
Toronto Declaration Amnesty Intl. + Access Now Declaration Machine learning, equality, and non-discrimination — a civil-society anchor for the Nelson framework.

Prompt for the room: which of these has any practical effect on what a public health researcher in Arizona actually does?

APH&AI · AI & the Earth
Part 03 · Open Mic
Part 03

Over to
the room.

Seed prompts, not a script — the framing above is shared substrate, and disagreement with it is fair game. The mic is yours.

APH&AI · AI & the Earth
18 / Questions
Six Seed Prompts For The Room

Pick the one you most want to argue with first.

Q1 · Footprint
Is "use a smaller model when you can" an adequate environmental ethic — or is it a personal-virtue dodge?

Compare to "use less plastic."

Q2 · Trade-off
When does a documented public-health benefit of an AI tool outweigh its documented environmental cost — and who gets to decide?

The hardest question. Save it for last.

Q3 · Legitimacy
If corporate AI constitutions are structurally insufficient, what is the minimum public mechanism that is sufficient?

Not principles. Mechanism.

Q4 · Civic grammar
Is the convergence across partisan lines a sign of strength — or evidence the vocabulary has been drained of force?

Nelson asks both. Pick one to defend.

Q5 · Magnifica Humanitas
Does the theological vocabulary (infinite dignity, transhumanism, posthumanism) help or hinder public deliberation about AI?

You can dislike the framing and still answer.

Q6 · Public health
What algorithmic harm in public health practice today would most benefit from a Marshall-style social-rights response?

Be specific. Name a system.

APH&AI · AI & the Earth
19 / Resources
Bookmarks

Where to keep reading after the room empties out.

Course page
tyson-swetnam.github.io/intro-gpt/ethics/
IEA · 2025
Energy and AI — data-center electricity to ~945 TWh by 2030 · iea.org
Li et al. · 2023
Making AI Less "Thirsty" — the secret water footprint of AI · arXiv:2304.03271
Patterson · 2021
Carbon Emissions and Large Neural Network Training · arXiv:2104.10350
Nelson · 2026
A civic grammar for AI rights, Science · DOI 10.1126/science.aeh7153
OSTP · 2022
Blueprint for an AI Bill of Rights · bidenwhitehouse.archives.gov/ostp/ai-bill-of-rights/
Leo XIV · 2026
Magnifica Humanitas · vatican.va/content/leo-xiv/…/magnifica-humanitas.html
More · intro-gpt
/environment/ · /bias/ · /legal/ · /transparency/
APH&AI · AI & the Earth
End · Thank You
Three days. One conversation that doesn't end here.

Keep
arguing.

The civic grammar gets traction because publics use it. The footprint shrinks because engineers and researchers refuse the default. Both are unfinished work — and both belong to the room that just spent three days in it.

Tyson Swetnam, PhD
tswetnam@arizona.edu · @tswetnam
BIO5 Institute · University of Arizona
PH&AI Summer School 2026