Self-Defined Society

An institutional theory of Redefinition Capitalism, autonomous value definition, and societal redefinition in the age of AI

​

Affiliation — a company name, a title, a degree — has long written our value for us. Generative AI is dismantling that institution: only in part, but for certain.

This paper is about the conditions under which that dismantling becomes freedom rather than despair. A society in which people write their own criteria of value and choose their own futures is what we call a Self-Defined Society.

VURA Working Paper No.6. This page explains the paper's core argument in plain English. The full paper (English original, 46 pages; Japanese edition, 40 pages) is linked at the end.

This paper in three minutes

• It discards one piece of received wisdom. "Quit the company and you will be free." Against the evidence, that premise fails.

• Unconditional money buys relief, not a new direction in life. In a trial paying $1,000 a month for three years, the largest use of the extra time was leisure.

• Affiliation was a bundle of three functions: identity, security, and execution. AI makes only the third cheap.

• So the central question is: who re-supplies the other two?

• Self-definition is a capacity, not a status. What counts is not whether you left an organization but how many alternatives you hold. A capacity can be scaffolded, measured, and distributed.

• The paper treats its own most convenient claims most harshly. The "one-person unicorn" was refused a proposition number.

The question this paper tries to answer

As AI makes the "execution" of knowledge work cheap, what remains scarce for humans? VURA's series keeps reaching one answer: the capacity to redefine what to build, and why. But the question cannot stop at the firm.

In a society where the company defined your role and the title defined your worth, what happens when that machinery loosens? Do people automatically become able to write their own value? If not, what is missing?

This paper takes on the social-structure (meso) layer of that question.

Starting point: dropping "quit and be free" on the evidence

The easiest answer is that affiliation itself is the constraint: leave, and autonomy follows. The paper rejects that route, because the evidence does not support it. Three counter-cases:

Counter-case 1: unconditional money bought relief, not reorientation

A US guaranteed-income randomized controlled trial (groups assigned by lottery) gave about 1,000 low-income adults $1,000 a month for three years, no strings attached — $36,000 of unconditional freedom.

In year one, stress and food insecurity improved. The mental-health gains faded from year two. The largest use of the extra time was leisure — not education, job search, or community work. Interest in starting a business rose; actual business formation did not.

Finland's national basic-income experiment raised well-being but not employment. Swedish lottery winners showed higher life evaluation but no change in daily emotion.

Resources buy relief. They do not buy a new direction in life.

Counter-case 2: de-affiliation itself does harm

A German panel study (the same people followed over years) tracked about 24,000 people for 15 years. On average, the fall in life satisfaction caused by unemployment did not recover even after re-employment. People adapt to marriage, divorce, and widowhood. They do not adapt to losing a job.

A meta-analysis (a pooled estimate across studies) of 324 studies put the distress gap between unemployed and employed people at an effect size of 0.51 (standardized; 0.5 is "medium"). Psychological problems affected 34% of the unemployed versus 16% of the employed.

Lost income cannot explain this. Employment had supplied time structure, status, social contact, and collective purpose.

Affiliation has also collapsed at population scale: the "deaths of despair" among middle-aged white Americans. The paper adds a caveat: correcting for age composition flattens the overall trend, and regional differences track drug supply better than economic hardship.

Yet the gradient by education survives the corrections. Those hurt most had the thinnest scaffolds — education, networks, outside options. The paper calls this the gradient of despair.

Counter-case 3: "independence" leaves the structure of dependence in place

The gig economy (one-off work contracted through platforms) is a natural experiment in de-affiliation marketed as self-definition.

The decisive evidence is a study of 112 platform workers. Side-income earners with a main job or outside options experienced the mechanism as "freedom." Those who depended on the platform for their main income experienced the same mechanism as "insecurity and control."

Self-employment research agrees. Voluntary independence carries a satisfaction premium; necessity entrepreneurs report lower satisfaction across 70 countries, and moving from unemployment into self-employment carries no premium at all.

So what was affiliation? A bundle of three functions

Affiliation was not a constraint but an institutional technology that bundled several functions and sold them together.

Definition 1 | Affiliation An institution that bundles three functions into one membership: (i) the conferral of identity and legitimacy, (ii) security and belonging, (iii) task execution and coordination. Function (iii) has an informational component (who does what, when) and a governance component (who holds authority, who bears accountability).

An analogy: the broadband bundle

An analogy. A home broadband contract bundles the connection, device protection and discounts, and the email address everyone knows you by. Even if you only wanted the connection, you bought the whole bundle.

Affiliation works the same way. Anyone who needed security (function 2) had to accept, wholesale, the identity the category assigned (function 1) and the tasks it assigned (function 3). That is what the paper means by an externally defined society.

When a cheap connection appears, the protection and the address do not come with it. The cheap part can be switched; the rest you must source again yourself.

Bundling was not a pathology. It made large-scale cooperation in industrial society possible. Its one cost: the criteria of your value were written somewhere other than where your life is lived.

AI replaces only the third function: execution

By one estimate, about 80% of the US workforce could see at least 10% of its tasks affected by large language models, with exposure rising with wage and education — the reverse of past automation.

But AI replaces only the execution and informational-coordination side of function 3.

The governance component — allocating authority and accountability — is not replaced. It is "exposed." It was invisible while fused with a supervisor's oversight; now it stands alone. Functions 1 (identity) and 2 (security) are neither replaced nor supplied.

The central problem is who re-supplies (i) and (ii), and how.

Execution is democratized; the gap in judgment widens

The evidence has two faces, and the argument depends on both.

On "framed tasks" with clear quality standards, AI compresses ability gaps. Across 5,172 customer-support agents, productivity rose about 15%, concentrated among novices. A pre-registered experiment (analysis plan filed in advance) with 453 professionals cut writing time by about 40%, the weakest writers gaining most. A field experiment in software development raised completed tasks by about 26%, mostly among juniors.

On "unframed judgment tasks," the sign flips. Kenyan small-business owners randomly given a GPT-4-based advisor showed an average effect of zero: the top half improved performance by more than 15%; the bottom half got about 10% worse, having acted on advice that did not fit them.

Among 758 consultants, tasks designed to sit outside AI's capability boundary produced 19 percentage points more wrong answers. And the user cannot see where that boundary is.

Economists split. Optimists see AI rebuilding a middle tier of "mass expertise"; skeptics put the ten-year cumulative productivity boost at 0.5–0.7%. The paper needs only what both accept: AI performs formalized execution and informational coordination. It does not define purpose, evaluate outcomes, or hold the authority to stop.

If a shortage of tools had been the constraint, business formation would have risen as access expanded — computers, open information, cloud, generative AI. The data show the opposite. US startup rates, job reallocation, and high-growth young firms have declined for decades while revenue concentrated in superstar firms. Inputs to capability rose; choices exercised fell.

What self-definition is: a capacity, not a status

Definition 3 | Self-Definition Not the status of having left an organization, but a capacity conditional on the structure of dependence. How many alternatives do you have? What does switching cost? Who sets the terms of the relationships you cannot exit? These divide outcomes.

In plain terms — Being self-defined is not about the business card. It is about whether you can walk away.

An employee who can write the purpose of her own role is far more self-defined than an "independent" worker whose prices, tasks, and display ranking are held by an algorithm. Leaving the organization is neither necessary nor sufficient. And a capacity, unlike a status, can be scaffolded, measured, and distributed.

What produces transitions is not resources but institutional scaffolds

From interventions that actually produced transitions — lump-sum grants in Kenya, goal-setting interventions, job-search programs for the unemployed, behavioral activation therapy, bridge employment — the paper extracts five shared features:

1. Articulation of purpose — writing, in your own words, what you want to realize

2. Staging — breaking a large transition into executable steps

3. A social setting — proceeding where others can see you

4. Commitment devices — designing, yourself, the cost of turning back

5. Inoculation against setbacks — building in, ahead of time, that failure will happen

The paper calls these the institutions of self-definition. They cost orders of magnitude less than unconditional transfers.

The "one-person unicorn": the claim refused a proposition number

The most seductive story about autonomy in the AI age is that one person will soon build a unicorn. The paper gave it no proposition number.

As of August 2026, there are zero verified cases. The story originates in a 2024 remark by an investor — literally a "bet." The most-cited example (solo-owned, $80 million valuation, six months) was in fact a company with eight employees. About 20% of unicorns were solo-founded by one tally, but that ratio predates AI, and all of them organized afterwards.

The macro record shows bifurcation instead: non-employer businesses are 78.4% of all businesses but 6.4% of GDP, while revenue concentrates in superstar firms. And the Kenya experiment showed AI widening the distribution of outcomes.

Proposition 8 | Bifurcation Solo entry and superstar concentration advance simultaneously.

In plain terms — The era when one person can start, and the era when a handful take everything, are the same era.

Proposition 9 | Frontier shift and widening dispersion The value frontier of very small teams shifts upward, and the dispersion of outcomes widens at the same time.

In plain terms — A small team's "ceiling" rises. So does the gap between winners and losers. This is not about the average going up.

The strong form — "democratized solo mega-value" — was refused a proposition number for lack of evidence (§7.4). It is the claim most convenient to VURA's own business, and the discipline of treating convenient claims most harshly is applied exactly here.

What a Self-Defined Society is: three tests

A Self-Defined Society (SDS) is neither a prophecy nor a eulogy. It is a specification, identified by three institutional tests.

Definition 4 | Self-Defined Society Test 1 | Distributed authorship — the capacity to write one's own criteria of value and purposes of activity is distributed widely through institutional scaffolds, not held as the privilege of the resource-rich Test 2 | Security without a price tag — security and belonging are supplied without demanding identity conformity as payment Test 3 | Coordination without surrender — coordination of joint work by organizations, platforms, or AI does not make giving up authorship of purpose a condition of participation

In plain terms — (1) Writing your own purpose is not reserved for the well-off. (2) You do not pay for security by "acting the part." (3) You do not hand over your purpose in order to work with others.

Test 3 does not forbid delegating means. Where purposes overlap, means are delegated by agreement; purposes are never handed over. Where there is no overlap, you can leave without losing security.

Three institutional mechanisms

First, extending purpose-defined roles beyond the firm (Proposition 10). Defining a role by the future to be realized and the principles of judgment, not a list of tasks (Purpose Description), extends to government and community roles.

But there is a shadow. Experiments show that overworking passionate workers is easily legitimized as "what they wanted." Spread the language of purpose without the security of Test 2, and society merely acquires a new tax called "passion." This condition is built into Proposition 10 itself.

In plain terms — Unless "you can survive writing the wrong purpose" comes before "you may write your own purpose," the exploitation of meaning becomes institutionalized.

Second, defending time (Proposition 11). Wales's Well-being of Future Generations Act (2015) obliged public bodies to consider future generations; the ten-year legal review found the duties procedural, weakly enforceable, and drifting. The paper classes the four-day-week trials as grey evidence and builds nothing on them.

In plain terms — Time is not "granted." It is "defended" with budgets and legal force.

Third, the design of authority between humans and AI (Proposition 12).

Halt authority: monitoring subtracts value

Article 14 of the EU AI Act mandates human oversight of high-risk AI, down to a stop button and awareness of automation bias (accepting a machine's answer unquestioned). The aspiration is right. The evidence is unsparing.

A meta-analysis of 106 experiments found that on decision tasks the human–AI combination performed worse than the better of human alone or AI alone (effect size g ≈ −0.23; the minus sign means the combination is worse). Over-trust in automation survives training and expertise. AI explanation features actually increase acceptance of wrong recommendations.

Posting a human as permanent monitor risks "accountability laundering": it subtracts value, then shifts the blame for accidents onto the weakest link — the human.

One design survives this record: exception-based Halt Authority. The prototype is Toyota's andon cord. Anyone who sees an anomaly can pull it. Stopping is cheap, blameless, and followed by collective diagnosis. Humans are poor continuous monitors but adequate "interrupters" when halting is cheap and structurally supported.

The design conditions matter. Financial-market circuit breakers (which suspend trading in sharp falls) show that mistuned thresholds pull prices toward the trigger and accelerate the crash — China withdrew its breaker in January 2016, four trading days after introducing it — while thresholds too coarse never fire.

A halt must be parameterized in advance, carry restart rules, and publish its activation rate. A zero activation rate is not health; it is an alarm. Halting must be legally cheap — if individuals bear liability, no one pulls the cord — and paired with training to judge without AI, or it decays into ritual.

In plain terms — "A human is watching" is not a safety device. Only wiring that makes stopping cheap and blameless is.

Developed further in Human on the Loop.

H2 | Building in the backlash: Proposition 14 and Japan

Tearing the foundations of livelihood away from protective structures always generates a counter-movement demanding protection — Polanyi's "double movement," in which market expansion and society's defensive reaction advance as a pair. Contemporary data agree: groups stripped of status show cultural backlash, and its intensity tracks the loss of status, not the change in income.

Proposition 14 | The protective counter-movement A transition without the re-supply of security generates a counter-movement demanding protection. Test 2 is not a luxury of the transition but its precondition.

In plain terms — Backlash is not an accident. It is the scheduled result of a design that put security last.

Japan carries a specific implication. Under membership-based employment — organized around "becoming a member of the company" rather than around a job — employer, identity, security, and role are more densely interlocked than anywhere else in the world. Unbundling strands more security and identity per role, so the backlash potential is large. But the existing rails of security that could be redirected toward Test 2 are also thick. Japan can become this transition's most dangerous laboratory, or its most promising one.

The deepest asymmetry: national statistics do not measure the capacity to choose the future

Society is a virtuoso at measuring the past. GDP observes realized production; labor statistics observe the state of affiliation. But no national statistic anywhere observes the capacity to choose the future — the distributed capacity to write one's own criteria of value and convert options into businesses and undertakings.

Proposition 13 | Observation asymmetry National statistics observe affiliation and realized production. They do not observe self-definition capacity.

In plain terms — We know every month how many people are employed. We have never once measured how many can write their own course.

So we cannot yet tell whether the long decline in startup rates is "an exhaustion of the will to redefine" or "an absence of the conditions for conversion." And what cannot be measured, capital markets cannot price. The informational precondition for Redefinition Capitalism has not yet been built.

The paper begins designing an indicator — as a research program, not a ranking — under eight constraints: answering the critique of composite indicators (several measures weighted into one number), mandatory sensitivity analysis, testing question-framing effects, triangulation with behavioral data, Goodhart resistance (a measure that becomes a target stops being a good measure), never summing capacity with outcomes, legitimacy in defining the "self," and honest display of lags.

And it writes this sentence about itself:

An instrument built single-handedly by a party that stands to profit from it fails the independence requirement.

What the paper admits it cannot yet show

The paper presents five definitions and fourteen propositions, each proposition carrying a falsification condition — the result that would show it wrong.

Proposition 2 (where de-affiliation fails), for example, is rejected if a low-capacity group flourishes without scaffolds, or a high-capacity group is harmed at the same rate — but capacity must be measured by antecedents fixed in advance; capacity inferred after the fact from outcomes is inadmissible. The escape hatch is closed in advance. Section 10.4 adds four conditions under which the theory as a whole is rejected.

On self-citation: earlier papers in the series are cited as axiomatic foundations, design proposals, or conceptual premises. The only observational weight they carry is the co-occurrence with defended time in the 18-company case study plus 5 controls, noted as unmatched. No proposition depends evidentially on a claim specific to the series.

And it discloses a structural conflict of interest.

Before this paper's theory existed, a press release of 27 May 2026 had announced an implementation-support business for the "Self-Defined Society." The suspicion that the theory was written to ratify the business is legitimate.

The response: the most restrictive treatment for the two claims most favorable to the business — the one-person-company argument (Propositions 8 and 9) and the societal extension of purpose-defined roles (Proposition 10).

Implications for executives, individuals, and institution designers

For executives. How much "authorship" do your employees hold — how far can they write the purpose of their own role? Do you supply the identity and security affiliation used to provide deliberately, or by the inertia of the bundle? The more execution AI takes on, the more governance is exposed.

For working individuals. The question is not "stay or quit." It is the structure of dependence. How many alternatives do you hold? What does switching cost? Who sets the terms of the relationships you cannot exit?

For institution designers. Unconditional resources buy relief, not reorientation. What works is the five-element scaffold — at a cost orders of magnitude below the resources.