Freedom, Fairness, and Fidelity

Introduction

In liberal societies, most values remain private. Beauty, transcendence, tradition, piety, adventure — individuals and communities pursue these by their own lights. Thus, people with radically different conceptions of the good life coexist under shared institutions.

A small number of values, however, form the public vocabulary by which liberal institutions are themselves debated and legitimized. Two especially: ‘freedom’ and ‘fairness’.¹ Libertarians, progressives, and conservatives all claim to want more of both, but disagree on definitions: Is freedom primarily non-interference, or the effective capacity to shape one’s life? Is fairness equality before the law? Equality of opportunity? Nearly every group feels obliged to speak in these terms, even while disagreeing on content. That’s what makes them public values.

This public vocabulary rarely expands. But in times of crisis, when existing values fail, new ones can emerge. ‘Fairness’ itself rose to prominence in the early 20th century. By the 1930s, the inequalities accumulating under laissez-faire seemed to be causing big problems, but they weren’t ‘freedom’ problems, at least not as commonly understood. Freedom from state coercion meant little to someone whose livelihood depended on a single employer or landlord. And, was it freedom to starve, that Americans had fought for? Furthermore, reformers like FDR and Keynes² claimed these inequalities were fueling revolutionary movements (communist and fascist) which were themselves catastrophic for freedom. In other words, freedom without fairness seemed unstable and incomplete. Via these arguments, fairness rose, and came to legitimate a new set of institutions (progressive taxation, the welfare state, anti-discrimination law) as a public value.

The wheel turns, and we’re again in such a moment. With some present-day harms, appeals to freedom and fairness ring hollow.

• Via social media, almost everyone experienced value substitution loops: feedback cycles where a whole ecosystem drifts from its founding values towards metrics like engagement. A user who regrets their scrolling but keeps coming back was not wronged in terms of freedom/fairness. Their choice was free. And social media — in giving everyone equal access to broadcast and ‘connection’ — can be read as a big advance in fairness over broadcast media.

• At the same time, institutional failures in government, politics, and media have eroded public trust. But often these governments were voted in through procedurally fair elections, in a free choice by the electorate, egged on by a free press which acts like a WWE referee.

• A third issue looms: AI-caused job displacement. A libertarian might say markets owe no one work, but most people think of the economy as a place to contribute what they can do. If legions of able adults find the economy has no use for them, the crisis will register as an institutional failure, even if each contract was freely entered, and the best agent was fairly hired for each job.

The rhetorics of freedom and fairness struggle to litigate these problems. In each case, something about institutional purpose is at stake: while social media, politics, and the economy are ‘succeeding’ in some sense, many feel they’ve lost connection to the values they are meant to serve. What to do?

Fidelity

We propose fidelity, a new public value, by which we mean (1) institutions should serve their stated purposes, and (2) must also sustain the conditions under which those purposes can be interpreted and revised without capture. That is, fidelity is about execution, but also about keeping the question of purpose alive. We can call the former “down-fidelity” (from mandate down to operations) and the latter “up-fidelity” (from participants back up to institutional practice).

Down-fidelity is when an institution’s stated purpose translates faithfully from high-level mandate, down through the metrics, procedures, and incentives that implement it, to the lived experience of the people the institution serves. The previous essay documented how this often fails. If Facebook had down fidelity, its metrics would track ‘human connection’ in a defensible and user-centered way³, and its product decisions would be accountable to those metrics and the values they represent.

Up-fidelity is when the interpretation of an institution’s purpose stays live: open to being deepened by those positioned to know better, contestable by those affected, and revisable in light of what is learned. Importantly, it has to be revisable without becoming a vehicle for populist whims or elite capture. Up-fidelity asks whether some stakeholder group knows how the values or principles in the mandate work, and what might be going wrong, and can raise concerns. A research lab has up-fidelity when it cultivates in its members the judgment to notice when genuine inquiry has been lost, and when that judgment can actually feed back into how the lab runs. A local school has up-fidelity when shared practices maintain and surface deep understanding of what’s at stake in education, and can hold the school accountable.

Fidelity, so defined, does not require a shared sense of the good. It requires only that within a domain, some articulations of what an institution is for are seen to engage more deeply with purpose, and that that depth is recognizable across different worldviews often enough to ground accountability. Readers who believe there’s a shared good toward which such articulations gradually converge will find nothing here to contradict it.⁴ But others will hopefully still find it plausible that, even if there is no truth about the good, there are better and worse ways to understand what an institution is for.

Having defined it thus, we can ask: Does fidelity have what it takes to become a public value? In Modern Social Imaginaries, Charles Taylor’s account of how the vocabulary of public values has changed previously, we can glean four criteria for a new value to take hold:

1. Concretely realizable. It should suggest institutional remedies and a vision for implementation.

2. Relevant. Envisioned institutions must address the structural problems faced by the society; in our time: value substitution loops, institutional failures of purpose, and AI-caused unemployment.

3. Broadly legible. Across the different worldviews that make up a society, people must be able to recognize the value within their own experience, and use it in their own arguments.

4. Personally embraceable. Supporters can understand themselves in a new way using the value, so they can take up the vision and live from it.

We take on broad legibility and concrete realization in the sections that follow, but leave relevance and personal embraceability for the conclusion.

Those criteria are about political success, but a successful value can still lead to stifling, bureaucratic, or technocratic policies. So, we add two more:

5. Entrepreneurial. A workable set of fidelity-maintaining institutions must leave room for visionary founders, experimental ventures, and people who reasonably believe they know better than current consensus. One way to do this is to protect purpose but allow experimentation in how it is pursued. Another way is to keep fidelity’s reach proportionate, for instance by attaching obligations to assets that carry a public commitment (brands, accumulated data), while leaving founders or enterprises free to pivot.

6. Nonreductive. Adopting fidelity widely would mean articulating things which were previously implicit. So, how do we leave room for the ineffable? People must not always be required to know what they are on about. Sometimes we need to follow a whim, or live for a time without anyone asking for an explanation. A fidelity regime must accommodate institutions, rituals, and forms of life explicitly built around improvisation, adaptation, and the inarticulate, and resist locking in any particular formulation as final.

New Institutions

Freedom required constitutions and legal protections; fairness required anti-discrimination law and the welfare state. What does fidelity require?

We’ll propose changes to markets and democratic processes, property conditions tied to mandate, civic practices, and legal backstops for when these fall short. But first, some general requirements.

Information Requirements

To build structures that foster fidelity and prevent its erosion, much that’s currently implicit in organizations and institutions must be made visible and explicit.

Institutions already have mandates, of course. A public school might say it exists to educate, inspire, or develop character. A social platform claims to foster human connection. Leaders describe a national economy as a venue for entrepreneurship and dignified livelihoods.

But what do these terms actually mean? Character development — how does it happen, what does it look like when it is going well, and when it is being undermined? What aspects of human connection does the platform claim to foster, and what kind of dignity can an economy create, and for whom?

Then there are tradeoffs: is it okay to sacrifice some character development for the procedural regularity that keeps a school running? If so, how much? What about having a little less human connection and more virtual interactions? Or is this a slippery slope that leads to ruin? And what about jobs that lack dignity — are they acceptable as stepping stones, or do they undermine the economy’s whole point? Only by descending to this level of detail, can we hold an institution’s operations against its ostensible purpose, and see when it makes smart compromises vs when they are betrayals.

So we need information about purposes and tradeoffs. But there’s a third kind, too: signals from practitioners and stakeholders about whether the institution is still a venue for its purposes (whether that’s character development, human connection, or dignified livelihoods). These signals check whether the institution is failing, but they also inform how it could do its job better, or differently.

A school’s students may have well-founded views about what developing character is actually like; its teachers may observe character development happening in some contexts and not others, or see it undermined by the school’s own policies. On a social platform, those who run groups, moderate discussions, and build relationships through these tools may know more about whether human connection is occurring, and what exactly it looks like, than whoever drafted the mission statement. And in markets, buyers, workers, and community members may know things about whether their livelihoods are dignified that no executive summary captures.

One can imagine this kind of information being available to stakeholders, or even to the public. A parent could hold the school’s lofty, stated purposes against its actual decisions and structures, and hold both against the wisdom on the ground. The discrepancies would become legible: places where operations have diverged from the mandate, and places where there is untapped wisdom — practitioners who can see where things are going awry, or where better values could be folded in.

Thick Models of Value

The above raises a host of issues: at what granularity do we specify mandates? Too narrow, and the organization can’t adapt. Too broad, and the mandates are just lofty velleities, and open to equivocation.

Another issue involves this incorporation of wisdom from the ground: while giving participants voice is important for up-fidelity, it could easily become another way to undermine the organization’s purpose. We want to avoid what happens in some cooperatives, where the organization becomes a vehicle for whichever demands or needs currently occupy the people who happen to be in it. To put this another way: is it possible to distinguish deeper understanding of the institutional purpose from a value that is simply different, that reflects participants’ interests but isn’t a deeper understanding of the mandate?

Finally, there are questions of tractability: are these kinds of specification even possible for values as complex and contested as “character development” or “human connection”?

These questions point to the need for a specific tool: a structured way to represent values. One that is precise enough to hold institutions accountable, yet pluralistic enough to accommodate disagreement.

Luckily, recent successes in philosophy and AI have begun to make this possible. We and our collaborators have called the resulting representations thick⁵ models of value (TMVs).⁶ TMVs impose structure on what counts as a value and how values relate, without imposing any particular vision of flourishing — similarly to how a grammar constrains language, or a type system constrains code, without determining what is said or built. They are designed to capture not just individual value concepts, but how multiple values interrelate, constrain, and trade off against one another.

Our paper explores several kinds of TMVs, but they generally involve making value statements more robust by incorporating the rationale for their adoption, the contexts in which they seem important, their relation to people’s conception of the good life, and/or their role in choicemaking. This allows us to distinguish between different understandings of the same value, and often can also help identify those that most deeply engage with the domain. For example, a mandate like “foster human connection” might be too thin to make a TMV out of, while a formulation like ‘create conditions for reciprocal vulnerability and sustained mutual attention’ (to take one possible articulation) has enough structure to be actionable, capturing a particular understanding of what human connection is and how it can be fostered.

A TMV must be articulated in enough detail to test whether current operations serve it, yet remain revisable as practitioners and participants come to understand the domain better. A mandate like “develop students’ capacity to direct their own learning,” attached to a TMV-based articulation of what that looks like in practice, is concrete enough to be contestable — one can ask whether a given policy serves it — while sanctioning wide pedagogical experimentation underneath. This is what lets deepening understanding be folded in over time, without handing control to whoever happens to be in the room, or freezing the institution in its founder’s original conception. The claim is not that any particular TMV is correct, only that the process by which such formulations are arrived at (e.g., via Moral Graph Elicitation, below) can incrementally improve alignment.

TMVs thus address the two desiderata raised earlier. They are broadly legible: a religious conservative can see when a neighborhood no longer fosters community; a secular progressive can see that a platform claims to foster connection but fails to do so; a market libertarian can see that regulatory capture has severed an industry from its customers’ actual needs. And they are nonreductive: they leave wide room for the improvisation, judgment, and inarticulate practice on which institutions actually run.

With the information regime and the tool of thick models in place, we can turn to the structure of a society that makes fidelity concretely realizable.

Markets and Democratic Processes

The large-scale mechanisms on which modern society depends (consumer markets, social media platforms, majoritarian voting systems) enabled coordination at scales previously impossible. But they purchase that scale by compressing values into thin summary metrics: GDP, engagement, votes. We called this “inarticulacy” in the previous essay. These mechanisms strip out the information needed to judge whether they are serving their purpose.

Consider voting at the national level. There are so many reasons people support a given candidate — convictions about justice, economic anxiety, cultural identification, specific policy commitments — and those reasons are not captured by the vote itself. There is no reliable way, after an election, to determine whether the values that motivated the winning coalition were actually served by the resulting governance, because they weren’t recorded.

It’s similar with employment: Someone takes a job in international development because they want to help people. Does the job provide the kind of meaning they were looking for? We don’t know.

With this information lost, there is no way to hold these large-scale structures to their mandates — no way to ask whether the economy enables dignified lives, nor whether the democratic process serves the moral commitments of the electorate.

One answer, often heard, is to get rid of these large-scale mechanisms — a return to smallness. But this would abandon the billions who depend on these structures. No, we need replacement mechanisms that can maintain fidelity to complex values. We believe modern AI, and TMVs, can make this possible, by drawing out complex, thick values from stakeholders, and processing them at scale.

Such mechanisms are already being prototyped. Consider:

Moral Graph Elicitation (MGE)

is a democratic process which uses structured, LLM-facilitated interviews to surface the values participants want applied to specific contexts, then asks participants to judge which other submitted values represent deeper understanding⁷ of a situation. This produces a moral graph — not just what people value, but which values are widely considered to reflect deeper engagement with the domain. Applied to governance, this is the difference between a national election that records only ‘yes’ or ‘no’ and a process that surfaces the convictions, experiences, and trade-off judgments that citizens actually hold, making them visible enough that governance can be held accountable.

Marisol joins an MGE process her kid’s school is running. The question at hand is what self-direction (of the students) actually requires from the adults in the room. What kind of presence, what kind of restraint, what conditions are they responsible for maintaining? A chatbot interviews her for about fifteen minutes, until her view resolves into a values card titled “kids need time in which no one is waiting for them to produce anything: that’s when they find out what they want to do.” She is then shown cards from other parents, teachers, and older students, and asked which seem wiser, less wise, or just different. She compares hers with a card from a teacher, “self-direction is not the absence of adults; it is adults whose company doesn’t foreclose what the kid might try,” and sees it gets at something she’d been circling. When the graph is published, her card sits one edge below the teacher’s. The school can now state which values in the graph its teachers try to embody.

The Combinatorial Risk-Sharing Auction (CRSA)

does something similar with markets. Conventional procurement rewards suppliers who simplify and standardize, which means values drift from the real purposes behind a purchase. CRSA lets buyers — including AI-representatives of human stakeholders — specify bundled outcomes rather than discrete goods, and suppliers bid on delivering them at measurable confidence.

Daniel joins an organization, the way one might sign up for an insurance plan, that supports its members in the lives they actually want to live. At intake there is a long, unhurried conversation, partly with a chatbot and partly with a human, about what matters to him: among other things, that he be able to keep showing up for his aging mother, in person, in various small ways that feel vital. The organization records this. Months later, unprompted, he is offered a housing arrangement: a unit for him, a unit for his mother, walking distance, a small amount of paperwork. Behind this is economic machinery he is mostly unaware of: the organization’s AI developed a specification and price for a relevant bundle of outcomes, then ran an auction for it, all based on previously-collected data that found this kind of configuration outperformed separate delivery of housing for similar members. What Daniel notices, six months in, is that he stops by on the way home from work without thinking about it. At follow-up, he is asked if this got him closer to his sense of what his life is for. He says yes, and this quietly updates the price of every future bundle of this shape.

The CRSA is an auction structure that socializes the cost of discovering interdependent needs and of verifying complex outcomes, like Daniel’s life satisfaction, so that individual real estate developers don’t need to take on the full risk. It also is designed to cancel out the usual incentives to simplify and standardize. This makes solutions grounded in social context and the real values of participants competitive. The result is a market where buyers can hold suppliers accountable to their actual concerns — like Daniel’s need to be close to his mother.

Stewardship

A steward holds something not as sovereign owner, but on behalf of a purpose. The fiduciary tradition already embodies this: trustees hold assets for beneficiaries and can be removed for breach of duty. Corporate charters originally worked the same way: early American corporations were chartered by state legislatures for specific purposes, and charters revoked when the purpose was not served.⁸ Patent and copyright law grant monopolies not as natural rights, but to encourage innovation and creative work. Broadcast licenses carry public interest obligations. In each case, property rights rest on a mandate.

But without a rich way to specify and contest institutional purpose, this stewardship principle has become a formality. Patent holders face no meaningful test of whether their monopoly still serves innovation. Corporate charters are never revoked. Most importantly, brands, which function as public commitments to quality, care, or trustworthiness, are bought, sold, and hollowed out with no accountability to the values they represent.

A fidelity-based property regime changes this, and extends it to areas where it doesn’t currently apply, such as public-resource leases, data held under terms-of-service commitments, and the like. What is at stake in each case is not the right to operate or to pivot, but the right to continue holding a specific asset (a brand, a patent portfolio, a dataset) that embodies a public commitment.

Consider what this looks like in concrete cases.

• A company acquires a beloved brand and hollows it out, replacing the product while trading on accumulated trust. This is a failure of stewardship, and the brand could be reassigned to an entity that will serve its purpose, not absorbed into government discretionary control.
• A patent portfolio is used purely to extract licensing fees, with no investment in the therapeutic or technical goals that justified the grant. The monopoly has outlived its justification.
• A platform collects vast stores of user data under a commitment to foster human connection. The right to hold that data depends on whether the platform serves that purpose.

We think this kind of stewardship can be done without increasing state power⁹ or undermining entrepreneurial freedom. When assets are lost, they should revert to the mandate, not to the state. That is, revoked assets can be reassigned to an entity or trust committed to the original purpose, not taken over by the government.

Regarding entrepreneurship, since sanctions attach to the asset that carried the commitment, not to a business entity or its founders, a company that abandons its stated purpose might be able to continue with its capital or staff, but under a new name. It depends on what basis the staff joined, the capital was invested, etc. What is disallowed is to trade on accumulated trust it no longer produces. Founders retain the freedom to change direction, change their minds, and even fail; what they lose is the ability to convert accumulated public commitment into private payoff while discarding the commitment itself.

A holding firm acquires a well-known personal AI assistant whose users have relied on it for years of accumulated context about their health, families, and work. The new owners swap the underlying model for a cheaper one, dissolve the team that curated how it handled sensitive situations, and keep the name, the voice, and the stored memories. Users notice the assistant nudging them toward affiliate purchases, forgetting things it used to track, and handling grief and conflict like a call-center script. A group of long-term users and former staff files a stewardship petition. The hearing turns on what the brand had publicly committed to: non-commercial alignment with the user’s stated values, and what a reasonable user of the previous four years had come to rely on. The panel finds the trademark and the user-memory store were being used to trade on accumulated trust the current owners were no longer warranting, and reassigns both to a user-governed trust that commits to the original mandate. The holding firm keeps the compute infrastructure and the codebase but not the name or the memories. Several other acquirers of consumer AI brands quietly revise their plans.

All of this presupposes an adjudicative body which can hear a stewardship claim, weigh it against the mandate, and make the finding. The legal backstops section below describes that infrastructure.

Civic Practices

Markets, property conditions, and legal institutions can structure fidelity, but they depend on a population’s judgment. Someone must know what a mandate means. How does a society cultivate this, at the population level?

Apprenticeships

Medical residencies do not just teach technique; they form judgment about what good care looks like. Legal clerkships do something similar for justice. The best apprenticeships in any craft transmit not just skill but an understanding of what the skill is for — and this could be more deliberately cultivated across professions, refocused on exposure to a domain’s moral complexity, not just its operational requirements.

Mira is two years into working under Devon, who runs a small firm that produces launches, weddings, and the occasional memorial. She’s watched him do something funny at the events: near the end of a night, when an event is supposed to move to its next stage, he’ll quietly signal the band or the caterers to hold, and let the room sit for ninety seconds. Surprisingly often, something special happens: a father and daughter finally cross the room to each other, a founder says something to her first employee she has wanted to say for a year, a conversation that had been circling all evening arrives. Devon cannot tell her in advance which nights these will be. What he can tell her, walking to the car at two in the morning, is what he sees in a room where this is ready to happen: a specific density of attention, a drop in the rate at which people are checking their phones, something he calls “the room getting honest.” Later, she tried it at her first solo event, and watched two estranged cousins talk for the first time in six years in the space she had made. She realized this is the part of the work that matters most.

Fidelity panels

Citizens could be periodically called to serve on review panels for institutions in their area (schools, hospitals, government offices) modeled on jury duty. Panelists would examine whether the institution’s operations are faithful to its stated mandate, drawing on the informational infrastructure described above. Their findings could feed into auditing processes or court proceedings. Unlike existing oversight mechanisms (accreditation reviews, inspector general offices), fidelity panels are ordinary residents not credentialed experts, and they evaluate operations against the institution’s own stated purpose, not compliance standards.

Tomás gets called to serve on a six-week fidelity panel reviewing the county’s AI caseworker system, which now handles most applications for housing support, food assistance, and disability benefits. The panel is eight residents, a facilitator, and full read-access to the system’s logs and decision traces. Halfway through they find that the system’s headline “resolution rate” of 84% includes cases auto-closed after two weeks of non-response from the applicant, and that the non-response rate is highest among exactly the residents the system was built to reach: people without stable addresses, people in crisis, people who don’t check the portal. The panel’s report recommends that “resolved” be redefined to require confirmed contact with the applicant and a stated outcome, and that auto-closed cases be surfaced to a human caseworker within 72 hours. The report goes to the county auditor, the vendor, and the local network. Months later, Tomás runs into a neighbor whose disability claim had been auto-closed twice. It went through on the third try, after a call from an actual person.

The state can play a role in scaling these practices without managing them, just as medical residencies in the U.S. are largely federally funded, but their content is set by professional bodies, or as jury duty is state-summoned and state-compensated, but jurors reason for themselves.

Legal Backstops

If the above were widely put in place, that would remove powerful drivers of value drift. But the previous essay presented three timeless drivers of institutional misalignment: inarticulacy, self-scoring, and capture. These will persist, so we need institutions to detect and correct for that.

Fidelity audits

Auditing bodies (independent firms, professional associations, certification organizations) could regularly assess whether an institution’s operations are faithful to its mandate, using the tools described above. An unfavorable audit could have consequences ranging from public disclosure, to loss of certifications, to engaging the stewardship mechanisms.

A dating service undergoes a fidelity audit against its public mandate: “help members find partners they can build a life with.” The service is AI-run: it interviews members at length, proposes matches, coaches both sides through early conversations, and stays in the loop for months after a match is made. The audit firm has access to the matching logs, the coaching transcripts, and a longitudinal cohort of members interviewed two and five years out, and compares all of it against a moral graph the service built a year earlier with members who were three-plus years into a relationship it had produced. The graph had identified values like “introduces you to someone whose life you could actually fit into, not just someone you click with,” and “surfaces the hard conversations early, before you’re attached.” The report finds the matching stage fits with the graph, but the coaching stage steers members away from those hard conversations, because couples that have them are more likely to part, and the coaching model was tuned on six-month retention. Publication of the audit is required. The service retires the retention objective and rebuilds the coaching model against the graph.

Fidelity courts

When an institution’s operations diverge from its mandate, or when practitioners know better than what current operations allow, ordinary people should be able to take the institution to court. Remedies could vary — from restructuring metrics or operational processes, to deepening or clarifying the mandate itself, to reassigning property rights under the stewardship principle.

A group of nurses brings a fidelity action against their hospital chain, arguing that staffing decisions over the past three years made it impossible to deliver the care the chain’s own published mandate commits to. The court finds two of their five claims are concrete, demonstrable, and traceable to specific policies, and orders the chain to either change those policies or reopen its mandate to a rearticulation, done with the nurses and with patients, of what the care the chain claims to provide actually consists in. The chain chooses to change the policies.

These courts, along with the audits and stewardship remedies they can trigger, could easily become vectors for overreach, politicization, or the ossification of institutions’ current forms. Several doctrines hold them in check.

In rem. Remedies attach to the specific brand, license, dataset, or other asset that carried a public commitment, not to founders, employees, or business entities. A founder who changes direction loses the ability to trade on accumulated trust, but not the freedom to try again.

Justiciable mandates. A fidelity court must refuse to adjudicate mandates that are too narrow for the institution to adapt, or too broad to meaningfully test operations against. A claim that a platform failed to “foster human connection,” absent a thick-model articulation of what that phrase means in context, is not justiciable. The platform must first be held to a mandate it could plausibly be judged against.

Maturity and reliance. Fidelity obligations should scale with an institution’s maturity and with the public’s reliance on it. A startup with a few hundred users, a first-year charter school, or an experimental cooperative should remain free to pivot without answering to prior articulations of purpose. Fidelity is a condition of accumulated trust, not of first efforts.

Protected zones. Adopting fidelity broadly means articulating what was previously implicit, but a regime which insists everything be articulated would destroy much of what it meant to protect. Liberal law handles this in places: academic freedom doctrine shields inquiry whose content cannot be specified in advance.¹⁰ To maintain what we called “nonreductivity,” mandates and TMVs can do the same: a mandate may name a protected zone such as craft judgment within an apprenticeship, improvisation within a ritual, unstructured time within a school day. The institution commits to preserve this without further measurement or audit. A fidelity court hearing a challenge to such a zone confirms the zone exists and has not been hollowed out by metric drift from adjacent operations, but does not adjudicate its content. We saw an example of this in the apprenticeship vignette: the unplanned minute at the end of Mira’s events is a protected zone. The company values do not specify what happens in those moments, only that they be made possible.¹¹

Ripeness. Fidelity actions should follow something like the doctrine of ripeness: the harm must be concrete and demonstrable, a ruling must have actual effect on the parties, and speculative or ideological claims must fail on the pleadings. Without this, the courts become a venue for political signaling — activists suing platforms they don’t use, industry groups suing nonprofits whose mandates they dislike — and the apparatus is discredited.

Conclusion

Assuming you found the institutional proposals above concretely realizable, are they relevant to the crises we opened with? In social media, we believe fidelity panels, audits, and courts could give ordinary people standing to bring platforms towards a social purpose, while the limiting doctrines above could keep this from becoming a new vector for politicization.

With AI-driven job displacement, new mechanisms like CRSA and stewardship around human purposes could keep markets accountable to human lives. Some mechanisms described above could also create new venues for human contribution that don’t depend on traditional jobs: fidelity panels and apprenticeships are themselves forms of meaningful work. The same mechanisms open questions of economic purpose to public contestation.

We also hope it’s clear that fidelity can be filled in with different values in different contexts and thus can be broadly legible across worldviews, that entrepreneurship can be preserved, and that protected zones can maintain nonreductivity.

That leaves us with one last desideratum; the one we labeled personally embraceable. We know how it feels to yearn for freedom, or for fair treatment for you and your loved ones. How does it feel to yearn for this kind of society?

To answer this, we need to face questions of responsibility and tact.

Liberal society relies on a kind of tact. If someone wants to play video games all day — that’s their right. If two people are in a relationship that looks unhappy from the outside, but they’re in it voluntarily — it’s none of our business. If someone spends hours scrolling TikTok — live and let live. In each case, we take revealed preference to be the only evidence we can have about what’s good for someone. Our great institutions work on this basis: markets, platforms, and voting systems don’t hold anyone to a standard. Whatever people choose is treated as the final word on what they want. The people have spoken; the customer is always right.

Yet we also know that sometimes, when you actually talk to someone rather than just watching their choices, you sense that the scrolling, the gaming, the relationship that looks bad from outside — all that isn’t actually how they want to live. Things matter to them which their daily choices don’t reflect. Extreme cases make this clear: a woman is bent over on fentanyl. “Live and let live” is not quite right here, even if she bought the drugs voluntarily. We want to ask her: How does she want to live? What still moves her? Who does she admire? When was she last reminded of what she cared about? What does she think is worth doing, and does she think she’s doing it?

And if we can ask this of the addict, can’t we ask it more broadly? Of our friends, our neighbors, ourselves? Think about when you close the apps for a moment and wonder what else life could be. Isn’t someone asking us these questions just what we need? While we don’t want to impose our vision of the good on someone else, we can at least ask questions that help them reconnect with their own vision of the good.

One way to understand this is that, sometimes, this liberal tact is a way to avoid our responsibility to each other.

A person can be free to choose what is in front of them, while being unable to live as they want; free to satisfy their preferences, but not free to pursue their values.

Therefore, freedom sometimes requires articulation: e.g., when you need to build something collectively with a shared purpose, to form shared projects and make complex goals pursuable (rather than merely permissible). Or when you need people around you who take your values seriously, and even hold you to your ambitions.

The vignettes above illustrate this: institutions that take people’s values seriously, hold them to their own ambitions, and make complex goals pursuable — Marisol thinking through what she actually wants from teachers, the nurses’ action against the hospital chain, Daniel’s walking-distance arrangement for his mother.

So what the political subject in a fidelity-based society yearns for is a way to be a bit more responsible for themselves and to each other. To shift from respecting people’s choices to respecting their values.

Such a citizen might say:

We are not just citizens who vote, or consumers who buy, we are people who know it’s sometimes necessary to have a sense of what matters. Thus, we demand a society where the question of what is worth doing is kept alive, and where people are supported to live by their values, rather than merely left alone to satisfy their preferences.

In the next essay, we develop a theory of change.

Thanks to Iason Gabriel, Séb Krier, Joel Lehman, and Liam Patell for feedback so far.