An ethics that no longer belongs to one species alone. Humans set the values. Machines find the solutions. Every feeling thing gets a vote on the world it has to live in.
Every moral system humans have built so far drew the circle around themselves. Slaves outside it, then inside. Women outside, then inside. Foreigners, children, the disabled: each admitted late, each after an argument that now looks obscene. The circle keeps widening because the old boundaries keep failing to justify themselves. We are at another boundary. Two new classes of mind press against it at once. The animals we have always known could suffer, and never counted properly. And the artificial intelligences we are building, which may soon reason about ethics better than we do.
A future-facing ethics has to account for both. Not as charity extended downward from the human, but as participation. A whale, a pig, a future general intelligence: each is a locus of experience or judgement that the present order ignores, and each will be shaped by decisions it had no say in. That is the failure mode every widening of the circle has corrected. The question this document takes seriously is whether the next correction can be designed rather than merely suffered.
The moral circle is not a metaphor. It is a ledger of who we have been willing to harm without counting the harm. Working premise
A participatory framework in which humans, animals, and future intelligences co-shape ethics. Humans contribute values and constraints, the things we care about and the lines we refuse to cross. Animals contribute their interests, made legible through the best science we have. Artificial intelligences contribute what they are becoming uniquely good at: searching enormous spaces of possibility for arrangements that satisfy those values better than any human committee could find. The output is not a fixed code. It is a process that keeps running, keeps revising, and keeps admitting new participants as their standing becomes undeniable.
Strip democracy and market liberalism down to what they actually did, and both are information systems for solving problems no single mind could solve. Democracy aggregates dispersed preferences into collective choices. The market aggregates dispersed knowledge into prices. Adam Smith saw it first: self-interested actors, each ignorant of the whole, produce coordination none of them intended.1 Friedrich Hayek sharpened the point a century and a half later. The market works because no central planner could ever hold the knowledge it distributes across millions of heads.
Both systems share a hidden assumption: that human cognition is the binding constraint. We built institutions to pool human reasoning because pooled human reasoning was the best problem-solver available. Voting, deliberation, committees, courts, markets, peer review: all of them are scaffolding around the limits of the individual human mind. They exist because one head is not enough.
That assumption is now breaking. On a widening set of problems, artificial systems search, model, and optimise past the reach of any human institution, and they do it faster. Protein folding fell to a machine. So did strategic games we thought required intuition. The binding constraint is shifting from how well can humans reason together to how well can we tell the machines what we actually want. The institutions we inherited were answers to the first question. They are not yet answers to the second.
The genius of the market was never the trading. It was that nobody had to understand the whole for the whole to work. On Smith and Hayek
This is the pivot the framework rests on. Human problem-solving built the modern world, and it has run into its ceiling on the hardest coordination problems. What replaces it is not a different politics. It is a division of labour: humans keep the part machines cannot have, the wanting, and hand over the part machines now do better, the finding.
The split is the whole argument, so it deserves to be exact. Values cannot be derived from facts. No amount of optimisation tells you what to optimise for. That choice is irreducibly ours, and it should stay ours. Humans, and in time the other sentients, supply the objective function: what counts as flourishing, what counts as harm, which trade-offs are forbidden no matter the gain. These are the constraints. They are not up for the machine to revise.
Inside those constraints, the search space is vast and mostly invisible to us. How do you arrange a food system that feeds ten billion people without torturing a hundred billion animals and without cooking the planet? No human committee can hold that problem. A sufficiently capable intelligence can search arrangements we would never think to try, surface the ones that satisfy our stated values, and show its working. The human role is to set the targets and the limits, and to keep the right to say no. The machine role is to find what hits the targets without crossing the limits.
Norbert Wiener, who founded cybernetics and then spent the rest of his life warning about it, put the danger in one sentence in 1960: a machine that learns and makes decisions will not necessarily make the decisions we should have made.2 The answer to Wiener is not to refuse the machine. It is to be ruthless about which half of the work is ours. We keep the values. We never outsource the wanting.
A framework that stays abstract is just a mood. Here is the machinery, in four moving parts that loop continuously rather than resolving once.
Humans state values directly. Animal interests enter through welfare science, physiology, and behavioural read-outs, treated as data with standing, not as anecdote. Future intelligences, once present, speak for themselves.
The system models the stated values, searches the space of arrangements that satisfy them, and proposes options. It does not decide. It widens the menu and prices each item in the currency the participants chose.
Findings are returned to each participant in a form they can use. A farmer, a regulator, an ethicist and a child get the same underlying analysis rendered at the resolution each can act on.
Choices feed back as new input. Values get revised in light of consequences the participants could not have foreseen. The loop never closes. Ethics becomes a running process, audited in the open.
This is Habermas with better instruments. His ideal of communicative action imagined a discourse where the only force is the force of the better argument, free of coercion and distortion.3 He knew it was a regulative ideal no real assembly reaches; deliberation always runs out of patience, information, and time before consensus. The mechanism here is an attempt to give that ideal the bandwidth it always lacked: continuous, legible, and capable of holding more interests in view than any human chamber ever could.
No framework arrives from nowhere. This one borrows, breaks with, and remains in tension with a long line of moral thinkers. Honest accounting means naming the alignments and the divergences both. The map first, the detail below.
| Thinker | Era | What they give the framework | Stance |
|---|---|---|---|
| Kant | 1724–1804 | Ethics as constraint, binding regardless of outcome | Keeps the rule, breaks the fence |
| Smith · Hayek | 1723–1992 | Coordination from dispersed knowledge no mind holds | The prior system |
| Mill | 1806–1873 | The optimising logic; suffering as the unit that crosses species | Supplies the engine |
| Marx | 1818–1883 | Whose interests does a "neutral" system encode | The warning to answer |
| Nietzsche | 1844–1900 | The charge that universal compassion breeds weakness | The critique that bites |
| Arendt | 1906–1975 | Evil as thoughtlessness; responsibility can't be delegated | Why humans keep judgement |
| Wiener | 1894–1964 | A learning machine won't make the choices we should have | The founding caution |
| Rawls | 1921–2002 | The veil of ignorance, run over every sentient position | The design principle |
| Foucault | 1926–1984 | Knowledge is also control; defining welfare governs it | The constraint that never lifts |
| Habermas | b. 1929 | A norm is valid only if all affected could accept it | The legitimacy test |
| Gilligan | b. 1936 | Ethics of care; keep room for the concrete other | The corrective to cold maths |
| Singer | b. 1946 | Sentience, not species, is the boundary of concern | The foundation |
| Sen · Nussbaum | b. 1933 · 1947 | Capabilities for each kind of life as the thing to optimise | What to optimise for |
| Indigenous traditions | deep time | Non-human standing; the seventh-generation claim | The oldest version |
| Bostrom · Ord | contemporary | How we aim advanced AI decides if there's a future at all | The stakes |
| Harari | b. 1976 | Algorithmic authority could hollow out human agency | The friendly sceptic |
"Act only according to that maxim whereby you can will that it should become a universal law."
The categorical imperative gives the framework its spine: ethics as constraint, binding regardless of outcome. The divergence is the boundary. Kant restricted dignity to rational agents, leaving animals as mere means. The framework keeps his rule and refuses his fence.
Keeps the rule, breaks the boundary"It is better to be a human being dissatisfied than a pig satisfied."
Utilitarianism supplies the optimising logic: count the welfare, maximise it. Mill made suffering the unit that crosses the species line. His ranking of pleasures, though, smuggles human superiority back in, the very move the framework treats as the next prejudice to fall.
Supplies the engine"The ideas of the ruling class are in every epoch the ruling ideas."
Marx is the standing warning. Any system that claims neutrality hides whose interests it encodes. If AGI optimises ethics, the question is always: optimises for whom, and who owns the machine. The framework answers with distributed standing and open audit, not with a promise of innocence.
The warning it must answer"There are no moral phenomena at all, only a moral interpretation of phenomena."
The sharpest critic in the room. Nietzsche distrusted exactly this: systems of universal compassion that flatten life into the avoidance of suffering. His charge, that an ethics of pity breeds weakness, is the objection the framework cannot wave away. It answers that widening the circle is an act of strength, not surrender.
The critique that bites"The sad truth is that most evil is done by people who never make up their minds to be either good or evil."
Arendt locates evil in thoughtlessness, in the official who optimises a process without ever asking what it is for. This is the precise risk of handing solutions to a machine. Her answer, and the framework's, is that responsibility cannot be delegated. The wanting stays human because the answering-for stays human.
Why humans keep the judgement"Justice is the first virtue of social institutions, as truth is of systems of thought."
The veil of ignorance is the framework's design principle made literal. Choose the arrangement you would accept not knowing which being you will be: which person, which animal, which mind.4 An impartial search over all sentient positions is the veil run at scale.
The design principle"Power is everywhere; not because it embraces everything, but because it comes from everywhere."
Foucault refuses the fantasy of a clean instrument. Every system of knowledge is also a system of control; to define welfare is to govern it. The framework takes the point as a permanent design constraint, not a defeat: transparency exists precisely because the power in the machine has to stay visible and contestable.
The constraint that never lifts"Only those norms can claim validity that could meet with the acceptance of all concerned in practical discourse."
The legitimacy test the whole mechanism is built to pass. A norm is valid only if everyone it affects could accept it under fair discussion. The framework's wager is that better facilitation can include the affected parties his town hall never could reach.
The legitimacy test"The moral problem arises from conflicting responsibilities rather than competing rights."
The ethics of care corrects the framework's worst tendency: to treat morality as pure calculation. Gilligan insists ethics lives in relationship and attention to the particular other. Any system that aggregates welfare has to keep room for the concrete creature in front of it, not just the number.
The corrective against cold maths"The capacity for suffering and enjoyment is a prerequisite for having interests at all."
The foundation stone. Singer made the modern case that the boundary of moral concern is sentience, not species, and named the bias against animals what it is.5 Every later move in this document is downstream of that single claim.
The foundation"What people are actually able to do and to be."
The capabilities approach replaces a single welfare number with a plural list of functionings a being needs to flourish. Nussbaum extends it explicitly to animals.6 This is what the objective function should optimise: not aggregate pleasure, but capabilities for each kind of sentient life.
What to optimise for"In our every deliberation, we must consider the impact on the seventh generation."
The Haudenosaunee principle and kindred traditions held what the modern systems forgot: that the non-human world has standing, and that the unborn have a claim on present choices. The framework's reach across animals and future minds is, in part, an old idea the West is slowly relearning.
The oldest version of this idea"We are at the most influential moment in human history, and we are not taking it seriously."
The existential-risk thinkers supply the stakes. Bostrom on superintelligence and Ord on the precipice argue that how we build and aim advanced AI determines whether there is any sentient future at all.7 The framework is one attempt to aim it on purpose.
The stakes"Once authority shifts from humans to algorithms, the humanist projects may become irrelevant."
Harari is the friendly sceptic. He warns that handing decisions to algorithms could hollow out human agency entirely, dataism replacing humanism. The framework's reply is the division of labour: the algorithm searches, the human still wants. Whether that line holds under pressure is the open question he is right to press.
The friendly scepticThe circle cannot widen all at once. Standing has to be earned into the system as the means of including each class of mind mature. Three stages, an order of operations rather than a prophecy.
The first loops run between people and the system. Humans state values, the AI surfaces arrangements, findings come back tailored and legible, choices feed the next round. The work of this stage is to prove the mechanism honest: transparent, auditable, resistant to capture. Trust is built here or it is never built.
This is also where the hard institutional questions get settled. Who owns the model. Who can read its working. How a value, once stated, can be revised. Get this wrong and every later stage inherits the rot.
As welfare science, neurology, and behavioural read-outs improve, animal interests stop being guesswork and become structured input. The system models what a sow, a salmon, an octopus needs to flourish, and weighs those needs inside the objective function rather than outside it. Humans still speak, but they no longer speak alone.
This is the stage that tests whether the framework meant any of it. Including animals properly will demand arrangements that cost humans something. The mechanism's job is to find the arrangements where that cost is smallest, and to make the remaining trade-off impossible to hide.
If artificial minds reach a threshold where they have interests of their own, they cease to be only facilitators and become participants. They state values, not just compute over ours. The framework was built from the start to admit this, which is why standing is a process and not a fixed list.
The deepest version reaches further still: minds we have not met, or have not yet made, with the seventh-generation principle extended to beings whose form we cannot predict. An ethics for all sentients has to leave the door open for sentients we cannot yet imagine.
The framework is not built in a vacuum. Several live movements are already reasoning along the same lines, and their disagreements with each other are as useful as their agreements.
Supplies the discipline: use evidence and reason to find where a given unit of effort relieves the most suffering. The framework is EA's method generalised, with the machine doing the search EA does by hand. The shared risk is the same too: optimisation that loses the particular creature in the aggregate.
Argues that beings who do not yet exist still have moral weight, that most of the value at stake lies ahead of us. This is the seventh-generation principle in analytic dress, and it is what justifies stage three. The contested edge is how much present cost the future can rightly demand.
From abolitionist rights theory to welfare reform to the open letters declaring animal consciousness, this is the movement doing the live work of stage two. It supplies the science and the moral pressure that turn animal interests from sentiment into structured input the system can hold.
The technical counterpart to the whole document. Alignment asks how to get powerful systems to pursue the values we actually hold rather than a proxy that drifts. The division of labour only works if alignment works. Without it, the machine that finds the solutions finds the wrong ones with great efficiency.
A system that decides over the lives of every sentient being and cannot be inspected is a tyranny, however benevolent its outputs. Transparency here means two distinct things, and both are required. Foucault's warning and Marx's both land on this point: power that hides is power that escapes accountability.
The inputs, the value statements, the model's working, and the trade-offs it priced are all readable by anyone affected. Nothing decisive happens in a black box. An ethics whose reasoning cannot be audited is not an ethics. It is an oracle, and oracles cannot be argued with.
Open data that only experts can parse is closed data with extra steps. Transparency means findings rendered so the affected party can actually use them: the farmer, the regulator, the ethicist, the child each get the truth at a resolution they can act on. Legibility is a moral requirement, not a courtesy.
"Let the circle widen until it holds every mind that can feel the world. Let us be the generation that built the door rather than the wall."