Telocracy

Introduction

We struggle to imagine a society that deals well with powerful AI. We must keep trying, however, because without a positive vision we’re haplessly reacting to economic and geopolitical shocks. Current proposals, like Dario Amodei’s “Machines of Loving Grace,” leave important questions wide open:

1. Questions of meaning, purpose, and connection. One consequence of AI agents is that we spend time with them. Do we then spend less with each other? If AIs replace us in the workplace, will we lose some of our social relationships? our sense of purpose? As Clara Collier¹ observed recently, work is where many of us mix with those outside our class or scene, and earn respect for our skills. A new social vision should say how humans can continue to find purpose beyond status games, and mix across social divides.

2. Questions of the proper division of labor. As AIs do more work, they’ll also do more decision-making, depriving humans of some kinds of experience, learning, and skill-building. AIs will increasingly draft rules, settle disputes, and interpret whether values are being fulfilled, and may come to exceed humans at some of these tasks. A new social vision must somehow define what humans should still do, and how they will go on learning, growing, and exercising judgement.

3. Questions of power and influence. If AIs do more decision-making, how can we ensure that human interests remain central? We currently attempt to keep government aligned with human interests via elections, the rule of law, and other checks and balances, and businesses aligned with consumers via markets. But these are already under strain: government agencies and companies drift away from what they were set up for, and voters and customers often can’t correct it.² AI agents can pursue whatever metrics we give them tirelessly and faster than any human. When those metrics don’t capture what we care about, or are gamed, misalignment with human interests grows more severe. A new social vision must therefore explain how humans can retain influence over the institutions that shape their lives.

I’ll present a social vision that addresses these questions, drawing from a strain of political theory called associationalism.³ Associationalism is the idea that more decisions could be made in associations — groups that people choose to join to pursue some purpose together under rules they agree on, such as churches, unions, cooperatives, or clubs.⁴

Below, I will focus on associations that are bound to a purpose and held accountable to it, rather than following whatever their members want at a given moment, and where those purposes can be revised in some ways, but not others, such that the purpose sometimes might outlive the current members’ interest in it.⁵ And I will focus more on lightweight associations that are easier to find, form, and leave. Some will be much lighter than a church or a union, such as a pool of buyers who want the same outcome, or a group of parents who share activities among their children for a time. An individual might belong to many.

Associationalism has always had a clear appeal: instead of ceding authority to the “artificial intelligence” of the market’s price mechanism,⁶ or to distant state bureaucrats, people in associations, those who share a concern, can make decisions about it themselves.⁷

And with these lightweight, purposeful associations, there’s an additional benefit: if associations form easily around new purposes, life can become more purposeful without us all agreeing on any particular one. The upshot is, hopefully, that the kind of “consumer choices” we have today, where we pick among products, can be augmented by a new kind of “values choice,” where we select which purposes we want to be part of.

I’ll call a society where this happens telocratic.⁸ Telocracy doesn’t require powerful AI, but it helps a lot: AI can help people form associations who otherwise wouldn’t find each other,⁹ and can lower the time cost of being part of an association and a custodian of its purpose. AI can also hold associations accountable to their purposes, interpreting and enforcing them across a wide range of cases, allowing greater trust in these associations and avoiding their capture by sectional interests.¹⁰

I’ll first cover telocracy in broad terms, then offer concrete institutional and legal ideas, including new market structures for group buying and for forming associations, a way to move government work to AI agents that are monitored by purpose, and some more radical ideas, like automatically reallocating assets (such as compute) to align them with purposes.

Telocracy

I will focus on modern liberal societies, where others have argued we will face acute problems of meaning, division of labor, disempowerment, and power concentration due to AI.¹¹ Introducing telocracy there means changes like these:

• Telocratic governance. More institutions will have extensively articulated purposes or mandates, kind of like Claude’s Constitution,¹² which can evolve over time in some ways but not others. Unlike bylaws, these focus on the purpose of the institution, not on procedure or what it owes to stakeholders. They can operate at various levels of abstraction — for instance, parents who disagree about educational methods might still agree to a school that tries multiple approaches to learn which work.¹³ Legally, these “telocratic” organizations may be companies, government agencies, or nonprofits, but they are more tightly tied to their purposes, via AI-based channels of interpretation and enforcement. These are organizations that have safeguards and monitoring much like those that frontier AI models already have today, but applied to the organization itself.¹⁴

• Purpose-fit groups. In a telocracy, people are assisted in finding and forming groups with a potential shared purpose. AI can help match groups and purposes by varying either: first, purposes can be posed at various levels of abstraction, mentioned above, to find a group which could sustain them; second, populations can be found that match variations on more concrete purposes. Continuing the example of parents and schools, different parent groups can be matched to schools with different approaches. Thus, a greater diversity of purposes can be realized in society. This could raise the ceiling, sustaining more ambitious projects (especially those which don’t fit neatly into state or market provision), and also raise the floor by allowing AIs to help disorganized and disadvantaged groups find each other, based on a common shared purpose, and coordinate.¹⁵

• Moving power around. Many functions, like public schooling, social media, or nursing homes, are controlled by the state or the market, but exist uneasily in either. When public schools are run by the state, this gives power to state officials, who may not be ideal for making those decisions. But if schools were instead run by market actors, decisions would be made by owners and managers, who also may not be the right people to wield power. The traditional third option is civil society: churches, unions, charities, clubs, and other associations. But for a century or so, liberal societies have let the state and the market absorb much of what these associations used to do.¹⁶ Part of why¹⁷ is that associations were less accountable than state actors, and less dynamic and responsive to people’s needs than market actors. This made associations an overall worse choice for many functions, and so they were crowded out. Telocracy aims to reverse this, by giving associations a new way to be accountable to their purposes (telocratic governance), and creating a kind of automatic entrepreneurialism in this third sector (purpose-fit groups). This could give us a new place to put functions we’d like to be dynamic and accountable, such as social media, public schooling, and nursing homes.

Telocratic Governance

Imagine writing a ten-page constitution for an organization. It describes what the organization is for, what trade-offs are legitimate, and where the boundaries of its mission lie.

Until recently, it would be difficult to know how such a constitution would apply across the thousands of cases the organization might encounter. But AI changes this: a model can be given a natural-language constitution and trained or prompted to apply it across a large space of cases. One can generate thousands of hypothetical situations, including adversarial and boundary cases, and ask which actions the model regards as consistent with the constitution, which it rejects, and which remain uncertain.

Imagine such a constitution isn’t as vague as most current mission statements, but has values spelled out with uncanny precision, detailing which trade-offs are legitimate in operations, and more. (See Appendix A for a precise account of what these purpose documents must contain.)

And imagine this document can change. The purpose can be revised as the organization learns, and it can move between more abstract and more value-laden¹⁸ formulations, depending on how much trust the organization wants to earn. A book club might just have a few sentences about reading hard books whereas a school or hospital might spell out more.

The document can then be applied in many ways: it can monitor internal decision-making, sending alerts when decisions depart from it. It can constrain AI agents acting for the organization. It can supply standards for independent auditors, and provide evidence in external adjudication. It can even be used for external assessment of how the organization delivers on its purpose: if the document claims the organization exists to deliver or protect a particular value, AIs can interview those impacted and see if their experience bears that out.

This new technology could enable big changes in the law, and in how institutions are governed. Western law¹⁹ often assigns governance to specified persons and offices, based on their interests. Think shareholders, members, voters, principals, plaintiffs, and beneficiaries. In each case, the essential question is what some entitled person, member, or constituency may demand. Call this “stakeholder governance.”

In government agencies and large organizations, we find a second kind: “procedural governance.” Here, the question is if the right steps were followed. The hope is for everyone to get the same treatment, with no important considerations skipped.

A third alternative, comparatively less mature, is “telocratic governance.” Its animating question is whether an institution remains faithful to its articulated purpose, which supplies an independent standard to which the institution is held accountable.

In small communities, telocratic governance isn’t new. Benedictine monasteries operate on the Rule of St. Benedict, which, in 73 chapters, describes the purpose of the monastery and how to pursue it. At larger scale, charities, public-benefit corporations, and purpose-specific public bodies offer a primitive form of such governance,²⁰ but their purposes are usually a sentence or two, and the same people are interpreting them as are supposed to be following them.

A big question for telocratic governance is deciding when these purposes can change. We don’t want institutions ruled by the “dead hand”²¹ of their founders, but we also don’t want them captured by current stakeholders. Some principles can guide us: for instance, purposes can bind more tightly once outsiders rely on the institution in various ways (living commitments), and bind more loosely when it’s easy for people to walk away.

Even when purposes bind tightly, they should be revisable when the purpose has been achieved, or when it can be better served another way (legitimate deepening).

I will touch on various reasons why telocratic governance can be preferable to stakeholder or procedural governance, but a simple one is that it aligns better with our ordinary understanding of institutions. When a team founds a company to make spaceflight radically cheaper, or a biotechnology company to cure a particular disease, or a school to educate in a particular way, we don’t talk of these endeavors as generic pools of assets to be continuously optimized for the arbitrary interests of whoever controls them. Human beings create an institution to do something.

So there’s a mismatch between our folk theory of institutions and our legal theory.²⁴ In ordinary life, we treat an institution as an undertaking whose present participants are, in an important sense, its custodians. They inherit something they can revise, but not appropriate.

As I argued in “Value Drift,” this mismatch between the law and ordinary understanding causes misalignment between an institution and the values it is supposed to serve. It creates vectors for capture, and paves the way for value substitution loops — situations where the institution and participants co-evolve into a strategic equilibrium no one wants.

In ordinary life, we treat an institution as an undertaking whose present participants are, in an important sense, its custodians. They inherit something they can revise, but not appropriate.

AI makes this more pressing. Under stakeholder governance, the AI agents which act for an institution are tools of shareholders, or of a ministry. Under telocratic governance, agents answer to the institution’s mandate, just as human colleagues hopefully would. It also makes it easier for the AIs: rather than trying to shape outcomes at the level of individual requests, they can help shape the larger processes of the organization.²⁵ And finally, as I’ll cover later in Moving Power Around and Economic Growth, telocratic governance could make a huge difference not just in the drift of institutions, but in our overall economic trajectory.

Purpose-Fit Groups

Once we have a landscape of associations, each with an articulated purpose, we can ask which people might be interested in joining them. This is a matching problem.

Currently, we get a lot of help as individuals in finding suppliers for bilateral contracts: we have Amazon, various directories, online product search, personal shopping agents, and malls. Powerful AI is already being used to supercharge this, to help companies find more customers.

We get comparatively less help finding each other, forming groups around shared purposes, and finding opportunities to act collectively. This second kind of matching problem is underdeveloped and underincentivized.²⁶ This is part of the worry that AI will further isolate us, each contracting via our own AI agents with suppliers, never discovering the people who share our concerns and could act with us.

A telocracy would find a way to incentivize this latter kind of matching. One idea is that people should have a right — not just to form associations — but to help with finding the people they could form them with. In other words, people have a right to discovery of purpose-fit groups.

This could be done via public bounties to a new kind of entity: matching mechanisms which lead to lasting or effective telocratic groups.²⁹

While rudimentary mechanisms where people with a shared concern find one another already exist (such as subreddits, wikis, Meetup groups, and assurance contracts³⁰ like Kickstarter),³¹ AI could supercharge this, by interviewing people about commitments, concerns, and values they have which currently lack organizational representation. Patterns of deep agreement³² can be discovered. When those shared concerns can be realized better together, this purpose-fit group can be assembled into a collective agent. (I’ll say more about this in Associational Discovery.)

There is evidence that groups don’t just form whenever there would be interested people, but are bottlenecked by the supply of social entrepreneurs who can take initiative and recruit the first few members.³³ Powerful AI could give us an unlimited supply of these “social entrepreneurs,” and thus, a Cambrian explosion of new associations. Thus, disorganized and disadvantaged groups might for the first time gather around causes that currently are too diffuse or too niche to find expression. Thus, ambitious projects which currently lack an entrepreneur to lead them, or which don’t even make sense as a business, but could find a place in a purpose-fit group.

Moving Power Around

Surveying the West today, existing associations are not very powerful. We have neighborhood associations, professional associations, churches, and clubs, but the decisions that shape our lives (food, housing, schooling, health care, work) mostly happen elsewhere, determined by market or state actors.

This is something almost everyone is uncomfortable with. Our rules today put some functions in regulated markets, some in unregulated ones, some in independent nonprofits, and some in the state. But people often think the wrong decisions are being made, and the wrong people are making them. We mostly don’t want schools run by the market, because we don’t trust business owners and managers with them, so we go with state-run schools, but many doubt state officials are good at running them either. Similarly, we don’t think government officials would run tech companies well, so they are left to the market, but many are unhappy with how owners and managers run them.

Telocracy could allow us more options. Formally, an organization with purpose-bound property (see below) and telocratic governance might be a corporation or a nonprofit, but in terms of trustworthiness, transparency, and dynamism, it’s a new kind of thing.

If one day these institutions are common and have proven trustworthy, the rules about where functions belong could change. I will propose some new rules, and how they could redistribute power.

• In Louisiana, you need a state license to arrange flowers. Under telocratic governance, a licensing board would be held to fidelity to its mandate — usually about public safety.³⁴ Decisions that mostly shield existing florists would be overturned and the agency might be de-funded and replaced. (I’ll say more about this in Government Agencies and Democratic Mandates.)

• Schools would also be held to fidelity, and the telocratic principle of purpose-fit groups would require schools to form around educational purposes families share, not just geography.

• Similarly, under living commitments, when a social media platform comes to mediate young people’s friendships, it could not be designed principally to increase engagement and ad revenue.

• Under fidelity, the platform’s metrics would have to track commitments related to those friendships, and it would lose relevant assets, like its brand, user licenses, or compute, if it didn’t. (See Purpose-bound property.)

Such structures could make a huge difference in the overall trajectory of the economy. If AI takes over much of the economy, and consumer spending falls as a fraction, telocracy could even keep these additional layers of the economy answering to purposes people recognize (see Economic Growth). And, when coupled with the AI-powered discovery of purpose-fit groups, telocracy could make market actors more purposeful, which would allow more functions to safely move into the market.

New Institutions

To give a sense of what a telocratic society might look like, I’ll propose a few innovations: purpose-bound property and contracts, new market and matching structures, and changes in how government agencies operate. I’ve also put two more institution designs in appendices: a court system to adjudicate failures in fidelity and purpose-fit grouping, and a new kind of professional association that keeps people knowledgeable about what the AIs are doing.

Purpose-bound property

The idea of purpose-bound property is that someone who owns or manages an asset — such as a landlord, branding agency, or compute provider — can ask for purposes from everyone who makes use of them, and issues a special kind of contract which binds the asset to that purpose. This requires no new law or public arrangement, and, as we’ll see, can still leave room for entrepreneurial innovation and flexibility.

Versions of this already exist. Land can be deeded to a town “for the purpose of a school,” but revert to the giver if it isn’t used that way.³⁸ Conservation easements bind land like habitat or farmland, and a land trust monitors and enforces them. Basic conditions are placed on brands and use of compute.³⁹ But these usually cover just a basic use or a list of banned activities, because enforcement has to be coarse-grained, via litigation. Here, we suggest replacing that with an infrastructure in which AI models interpret mandates and monitor fidelity to them,⁴⁰ and there are more fine-grained options for auditing, appealing, and reassigning assets to new purposes.

One could have contracts revert if buyers/lessors deviate from their purpose, but organizations often start unsure of their purpose, or of which purposes will work out. They may need to try several. And, as we said earlier, purposes need to evolve legitimately even later, due to legitimate deepening and living commitments, above. They must not fossilize the original purpose.⁴¹ So let’s consider a more sophisticated model, where so long as an organization doesn’t use assets they wouldn’t have been given without their purpose, they can pivot to a new one,⁴² and where — even after they do have such assets — they can continue to shape their purpose within limits.

Say an organization has a dashboard where they can see whether what is happening with their business could require them to lock in their mandate or make changes to it. If the organization is a growing social network, and they’ve started to mediate their users’ relationships, living commitments could require them to take on values they’d rather not support — values about lasting friendships, or good community discourse, for instance. They might then choose: (a) to mediate relationships less, by opening up data so different messaging apps can share their social graph or by providing algorithmic choice; or (b) to accept these new obligations; or (c) to forfeit their purpose-bound property, such as the brand and the social graph, to an organization that will take them on. In case of forfeiture, assets could be reallocated through a chain of custody to another organization (pre-existing or newly founded) which can credibly claim to continue the purpose.⁴³

If purposes can shift at the beginning of a contract — and if we imagine general-purpose compute and other assets being leased this way — then the question arises: what purposes are acceptable? We would rather not have each cloud provider deciding what purposes are worthwhile.

One approach is to allow any plausibly positive-sum purpose at the early stage of a contract. This would give asset owners a way to support value generation in general, rather than particular conceptions of the good.

The value-neutrality of this arrangement differs from market neutrality: while market trades are assumed to be positive-sum for the two parties involved in the trade, they can be zero-sum, or worse, for everyone else: a warlord renting compute to run a drone strike is not creating value overall, even if both parties gain from the deal. Plausible positive-sumness blocks arms races in general: firms trying to file more motions than the other side, lobby harder, etc. Without such a criterion, a provider that rents to the highest bidder may find more and more capacity goes to agents working against each other.⁴⁶

Values-Rich Markets & Matching

Innovations in contracts are cool, but they won’t shape outcomes across the economy unless paired with innovations in markets and matching. Consumer markets are mainly (with exceptions like Kickstarter, health insurance, and apps like ClassPass) organized through repeated bilateral contracts between a seller and an individual consumer. This bilateral structure determines the structure of the contracts and the kinds of goods that can be sold.

It works for goods we each use separately, like a toaster or a haircut, but less well for things like a dinner with friends, a group of parents who trade off child care, or an apartment near your aging parent.⁴⁷ Economists call these relational or club goods.⁴⁸ It also works less well when suppliers need to coordinate, such as when caring for a parent at home, which involves housing, rides, meals, and medical visits.⁴⁹ Bilateral contracts disaggregate or misspecify needs like these.⁵⁰ Plus, over time, markets have a bias toward shallow provision: the easiest needs to specify, compare, price, and satisfy bilaterally are served, while deeper needs are left unaddressed.⁵¹ Suppliers win who simplify, standardize, and provide individual solutions, while complex and relational needs suffer.

Bilateral contracting and voting are like the first two bowls of porridge from the classic tale “Goldilocks.” If Goldilocks uses bilateral contracting, she buys a single bowl of porridge and eats it alone, since this kind of market often fails to process purposes that can only be realized together. If she uses mass voting, she can only choose one bowl of porridge which everyone must eat, and she has to run campaigns against those who want raisins on top, no porridge at all, or who eat their porridge cold.

Coming up with alternatives to voting and bilateral contracting is the domain of mechanism design. Goldilocks needs a mechanism that’s just right: one that surfaces purpose-fit groups instead of treating people as isolated individuals or defaulting to a whole-population compromise.

Mechanism designers have some experience here.⁵² For instance, the U.S. medical residency matching program used to match one member of a couple first and then limit the other to jobs in the same city. This often went badly for couples, and by the late 1970s many of them were finding jobs outside the match. Starting in 1983, couples could rank pairs of jobs instead.⁵³ Another example of a mechanism design for joint outcomes is German Baugruppen — groups of future neighbors who commission an apartment building together. Instead of each buying a unit from a developer, they get cheaper flats that are built around the relationships in their community, with shared gardens and common rooms.⁵⁴ And there are many more examples in the mechanism design and game theory literature: a hedonic game⁵⁵ is one in which players have preferences over who they are grouped with; coalition formation, more broadly, means finding groups that benefit from pursuing a strategy together; and synergy is when multiple goods or transactions (such as tickets for a group of friends to attend a concert together) are worth more when combined. Unfortunately, there are many more papers in this area than implementations, and these ideas have barely been applied to consumer markets.⁵⁶ This is likely because, as mentioned in Purpose-Fit Groups, these kinds of matching mechanisms are harder to monetize.

Goldilocks needs a mechanism that’s just right: one that surfaces purpose-fit groups instead of treating people as isolated individuals or defaulting to a whole-population compromise.

Joint Outcomes

One such mechanism is outcome-based contracts and impact bonds.⁵⁷ These attach payments to rich outcomes, rather than delivery of a service. But they are still designed for one buyer at a time. Instead, we can imagine a pool of buyers, represented by AI agents, and a mechanism that posts bounties for shared outcomes amongst them.⁵⁸

Even impact bonds have been limited by the difficulty of specifying outcomes that are both measurable and meaningful. AI could help by eliciting desires in ordinary language and identifying shared purposes across differently expressed wishes. This could be combined with automated procurement, where AI agents specify requirements, issue collective requests for proposals, compare suppliers’ proposals, and negotiate commitments among participants.

A key question, even with impact bonds, is how to de-risk supplier engagement when outcomes are highly personal and uncertain. This is especially true for relational goods, where the outcome depends on the actions of many people. Risk-sharing mechanisms like performance bonds and amortized insurance can pool this risk, so suppliers can be paid for how they do overall.⁵⁹

In markets with these kinds of intermediating structures, suppliers can be accountable to buyers’ actual concerns — like whether a widower has a social life, or a family has fun together. Group interests can be discovered and supported, all while retaining the profit motive that drives efficiency and innovation. Without such structures, the supplier who sells something modular to each person gets there first, faces less risk, and has an easier time raising capital.

But we can do even better.

Associational Discovery

Imagine a mechanism that knows about values and commitments across a population. Perhaps it runs broad surveys, looking for collective goals that lack representation. When it finds a cluster with a potentially shared purpose, it could take a sequence of steps⁶⁰ to test if these similarities could support a joint undertaking. First, it might invite people to join for a single, time-boxed group act,⁶¹ such as attending a meeting, running a quick policy experiment, or engaging in some kind of political expression. This would only happen if enough people opt-in.³⁰ If that goes well, commitment could ratchet up towards recurring practices, shared resources, and finally a mandate and purpose-bound property (Purpose-bound property). At each stage, people can revise the proposal, decline it, or proceed only if others do.

Economic Growth

Right now, about 68% of U.S. GDP is household consumption: people buying food, rent, health care, disposable goods, entertainment, and so on. Another 17% or so is government spending, on schools, roads, defense, and public administration, and about 18% is private investment, in buildings, equipment, software, and research.⁶² Household consumption keeps the economy closely tied to what people want, and much government spending is close to people too. But that could change if the AI part of the economy grows much faster than the human part.

Even now, segments of the economy diverge from human interests. Some investment ramps up proxies, like the speed of trading algorithms or the clicks on social media feeds. Ramping up these abstract measures is presumed to increase human well-being, but the more abstract the measure, the harder it is to tell.⁶³ Even in the part of the economy that’s counted as household consumption, a good share doesn’t reach us as goods at all: it covers marketing overhead, compliance, defensive corporate spending, and finance.⁶⁴ Again, it can be hard to tell whether these actually increase human well-being. Some is clearly zero- or negative-sum, like lobbying, litigation, and other arms races,⁶⁵ or the value substitution loops I described in “Value Drift.”

Autonomous corporations could both lower the percentage of GDP tied to household consumption, and also remove the human oversight that currently unwinds these loops. If so, we get the kind of economic growth Nick Land celebrated, where capital deployments and machine intelligence accelerate away from human ends.⁶⁶

Telocracy could help: under fidelity, proxy metrics like clicks or citations couldn’t crowd out recognizable purposes. Under epistemic vitality, people would remain in close enough contact with work to intervene when things go off track. More of our AI intelligence would be spent reasoning about mandates and values, not just driving up a number. And mechanisms like purpose-bound property would tie assets, including compute and capital, to purposes, so investment would less often chase abstract metrics and feed negative-sum loops.

To human eyes, this telocratic growth will seem much more like progress, recognizable as furthering various human-relevant purposes, like flourishing schools, cures, restored biodiversity, etc.

Government Agencies and Democratic Mandates

I mentioned that telocratic governance could be an alternative to procedural governance. The idea with procedural governance is to follow a set of rules and procedures to ensure fairness. This has been the best we could do, because it has been impossible to judge cases on their merits consistently, without bias.⁷⁰

But, if models can apply the same purpose consistently, across locations and populations, using reasoning anyone could check, that could allow a new type of impartiality which affords more dynamism and adapts to changing circumstances.⁷¹ This could increase trust in government agencies, which has fallen in many places, like in the U.S.⁷²

Once agencies are staffed by agents, or if human decisions are logged, the public could check an agency’s decisions against its purpose and the outcomes it was entrusted to produce.⁷³ A distrustful Congress or Parliament could also condition agency funding on fidelity to purpose.

If government AIs apply mandates as they work, this could be safer: future superintelligent agencies may vastly exceed human officials in prediction, coordination, and policy design. Telocracy could allow this superintelligence to serve the public good as expressed in broad purposes, not just executive interests or political settlements.

Values-Based Deliberations

Government agencies (and some other associations⁷⁴) may want to update their mandates based on values and purposes that are widely shared. In this case, there needs to be a way to gather up public wisdom related to one purpose or another. Citizens, affected populations, and others with relevant experience could help articulate the agency’s purpose, supply cases that test proposed formulations, and help revise mandates as experience accumulates. This public participation would be richer than commenting on a finished rule.⁷⁵

AI-aided democratic processes like Moral Graph Elicitation (MGE)⁷⁶ can help. In MGE, a large population’s values are collected via structured, LLM-facilitated interviews, surfacing the convictions, experiences, and trade-off judgments that citizens hold. The LLM asks participants to surface values they want applied to specific contexts, and also to judge others’ submissions for whether they represent a deeper understanding of a situation.

Conclusion

In the introduction, I brought up several requirements for a new social vision that current proposals fail to answer. Let’s review them in light of what’s been described.

How will humans still find purpose beyond status games, and mix across social divides? At their best, modern lives already contain many sites of purpose: a professional in a modern city might have a job that puts them on purposeful teams around the world, sing medieval music in a choir, and train in capoeira. Meanwhile, their partner prepares for a marathon, and their kid builds robots. Each family member belongs to several communities of purpose, with its own values and community, and is free to move between them.

Let that not be a relic of the age of work! If, under telocracy, there are many more sites of purpose, and it’s easy to find those that fit, we can preserve — and even enhance — that diversity. Associational discovery, epistemic communities, and joint-outcome markets can help people find and sustain the activities and relationships that give their lives meaning, so we still need each other.

What should humans still do, and how can we still learn, grow, and exercise judgement? Under the interpretive and auditing mechanisms of Appendix A, there’s a division of labor: AIs reason about values and actions on our behalf but they don’t substitute for human judgment about what those values should be.⁷⁷ Humans keep whatever practice they need to remain competent judges of the purposes they care about. (See Appendix C for how to sustain those practices.)

How will humans retain influence over the institutions that shape our lives? Telocracy offers an opinion about which kind of power to retain: the power to shape institutions and to understand, in detail, what goods we are pursuing. It also offers a variety of mechanisms to maintain that power: on the softer end, epistemic communities and deliberative processes keep humans in contact with work, decisions, and values. On the harder end, purpose-bound property and values-rich markets tie economic power to the purposes we care about, and make it difficult for institutions to drift away from them. There are other reasons, not developed here, to think that a telocracy might do well in global economic⁷⁸ and geopolitical⁷⁹ competition.

Telocracy offers an opinion about which kind of power to retain: the power to shape institutions and to understand, in detail, what goods we are pursuing.

For all these reasons, telocracy seems like a strong candidate for our AI future. But many are hungry for it also on an individual level: not everyone finds purpose and meaning in what modern life provides. Especially those who work in industries prone to value drift seem to be searching for something more. I believe this is what Wolf Tivy means when he writes about his peers “trapped in some kind of propaganda complex, wasting their lives working for effectively nothing.” Those of us who have been lucky enough to find purpose in our work may not realize how many people don’t enjoy the same privilege.

I can’t think of a better way to address that than telocracy, which — as I said in the beginning — gives citizens much more power to name their own purposes and to shape institutions that serve them.

Perhaps the most exciting thing telocracy offers is the chance to decide: What do we care about? What’s worth doing together?

Appendices: A. Specifying Mandates · B. Purpose Review · C. Epistemic Communities