Who writes the rules — corporations or publics? What does an honest accounting of the compute, the water, and the grid look like? An open conversation to close the school.
Tyson sets the question, the vocabulary, the stakes. The slides that follow are the shared substrate — not the lecture.
Three voices working through six prompts. Brief openings, then real conversation — disagreement welcome.
Open Q&A. You've been reading and prompting for three days — bring the harder question, not the safe one.
An outside-in question. Regulations, declarations, constitutions, bills of rights. Asks who has standing, who decides, who is harmed, who is accountable.
An inside-out question. Refusal, safety, fairness, alignment. Asks how the system itself behaves under pressure — what it says yes to and what it says no to.
Most of today's panel lives in the first column. Most of what AI companies talk about lives in the second. The gap between them is where most of the disagreement actually sits — Siau & Wang (2020).
A small group of scientists gathered at Dartmouth for a Summer Research Project on Artificial Intelligence. They coined the term. The field begins.
For the next seventy years AI lived mainly in science fiction and in a small community of industry researchers and academics quietly building the digital infrastructure the modern systems would need.
What changed in the last three years isn't the idea — it's the infrastructure. The questions waiting from 1956 arrived all at once.
Asimov's framing — robots must not harm humans, must obey, must protect themselves — is fiction, not law. But the deep instinct it encodes (prevent harm to humans, by design) shows up in every modern AI safety practice: refusal training, red-teaming, evals.
The point isn't to apply the laws. The point is that the field still asks Asimov's question first.
The Turing Trap warns that pushing AI toward imitation of humans tends to replace workers rather than augment them — driving wages down and concentrating economic and political power.
"More human-like" is not a self-evidently good design goal. The frame matters.
Corporate AI constitutions and public AI bills of rights are doing very different work. Alondra Nelson argues one of them is structurally insufficient.
The model developer's vision of how its own model should behave. Canonical example: a frontier-lab "constitution" document used in training.
Developed through public process. Canonical example: the White House OSTP Blueprint for an AI Bill of Rights, October 2022.
Source: Alondra Nelson, A civic grammar for AI rights, Science (2026). DOI 10.1126/science.aeh7153
A civic grammar — non-discrimination, transparency, data privacy, notice, human alternatives — that has been traveling across jurisdictions, partisan lines, and institutional contexts.
An American tradition — Patients', Consumer, Tenants', Workers', Taxpayers' Bills of Rights. The AI Bill of Rights template is spreading the same way.
Institutional diffusion among weakly related actors (Strang & Meyer). Conceptual, not relational. Connecticut Democrats, Oklahoma Republicans, and a national student-advocacy network can adopt the same vocabulary because each, separately, met the same algorithmic harm.
"Rights are never fully delivered at the moment of declaration. They are successively rearticulated by publics who attempt to hold institutions to commitments those institutions have not yet honored." — Nelson, paraphrasing Marshall.
Renewed attention to social theories of how technology, human experience, and social order are entangled. Weber on rationalization, Du Bois on technology and inequality, contemporary work on algorithmic governance.
The systems themselves merit study as social, political, and economic artifacts — not just engineering products. Training corpora, labor relations, ideological commitments, institutional effects.
AI capabilities may transform — or upend — the practice of social investigation itself: large-scale text analysis, multimodal pattern detection, conversational interviewing at scale. Worth critical scrutiny, not uncritical adoption.
Diagnostic use: when you read a piece of AI-ethics scholarship, ask which imperative it engages. That often clarifies what kind of argument is being made — and what kind of counter-argument would land.
May 2026. Pope Leo XIV's first encyclical places AI alongside the industrial revolution — and arrives at the same conclusion as Nelson from a very different starting point.
May 15, 2026 is the 135th anniversary of Pope Leo XIII's Rerum Novarum (1891) — the foundational encyclical of modern Catholic social teaching, written for the dignity of workers under the 19th-century industrial revolution. Leo XIV places AI explicitly alongside that disruption.
Pope Leo personally presented the encyclical at the Vatican, alongside a co-founder of a frontier AI lab. The first time a pontiff has presented an encyclical himself rather than delegating to cardinals — a signal that the conversation is dialogic, not purely adversarial.
The project of using technology to overcome biological limits — aging, mortality, embodiment. Leo XIV rejects the framing of human finitude as a problem to be solved.
The philosophical position that humans and machines are continuous, or that human distinctiveness is illusory. Named as an active "anti-human vision" embedded in contemporary AI development — not merely a speculative stance.
Withdraw AI from military applications and from purely economic interests. Direct it toward the common good. The framing is deliberately stark.
Stricter state and international regulation of AI companies. Not industry self-governance. The encyclical does not treat the two as substitutes.
"Every person of goodwill" — the same rhetorical move Laudato Si' (2015, on climate) and Antiqua et Nova (2025, on AI) used to seek ethical common ground beyond doctrinal lines.
Before we invent a new ethics for AI, an inventory: the declarations, principles, and conventions you are already operating under whether you read them or not.
| 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. |
Panel prompt: which of these has any practical effect on what a public health researcher in Arizona actually does next Tuesday?
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.
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.
Evaporative cooling consumes potable or industrial water — sited disproportionately in arid regions. The Southwest is a relevant case in this room.
Accelerator manufacturing, server build, building shell, fiber, copper. The carbon cost is paid before the first prompt — and is rarely on the accounting.
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).
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.)
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.
Collective, diffuse, opaque harms that older rights frameworks address only partially.
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.
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.
Six prompts. The framing above is shared substrate — disagreement with it is fair game.
Not principles. Mechanism.
Nelson asks both. Pick one to defend.
You can dislike the framing and still answer.
Be specific. Name a system.
Compare to "use less plastic."
The hardest question. Save it for last.
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.