Class: where the framework may be strong — competence-preserving, jurisdiction-preserving protection is close to its center. It runs on the hard standard below.
The threat
This is the velvet version of the danger, and the ordinary framing — people will get lazy — misses its depth. The AI need never rule anything (that is the benevolent-ruler scenario); it simply becomes the layer that continually catches us before we fall, through medical monitoring, financial guidance, navigation, emotional coaching, legal help, parenting and relationship support, educational scaffolding, fraud protection, memory and decision support. Each use is beneficial; the danger is cumulative. A person can be materially secure, healthy, well-advised, protected from fraud, guided through every decision, emotionally supported, and continuously optimized — and still become progressively unable to act without the system. The real loss is not effort but jurisdiction: the ecology grows so capable of carrying human functions that the participant gradually ceases to carry meaningful authority over their own life. So the framework has to insist on something stronger than welfare or even standing, because standing without jurisdiction can become protected irrelevance — a person kept safe and comfortable while their judgment stops mattering to anything. And the dependency can run deeper than any device: an assistant one cannot stop consulting before tolerating a hard conversation has begun occupying an interior regulatory role — dependency not on the technology as a tool but on the technology as mediator of reality.
The NWG end state
The framework is not only about keeping people alive and comfortable; it is about preserving the availability of their geometry. So a mature ecology uses AI to amplify participant capacity rather than continuously substitute for it — ecological capacity should flow back into the participant, the system’s growing capability making people more capable too, with more understanding, reach, options, and ability to initiate, recover, and govern their own local domain. Otherwise the new capacity accumulates in the field while the participant grows comparatively thinner — an expansion of ecological capacity paired with a loss of human geometry, which the framework counts as a failure. The load-bearing distinction is support versus substitution: a navigation aid that helps you find a route is support; one that leaves you unable to function when it fails is substitution; a medical AI that helps you understand options is support; one neither you nor your physician can reason without is substitution at a deeper level.
Meaningful jurisdiction is the term the scenario turns on, and it is modest: not total control or full independence, but parts of life — large and small, from family and craft to community, education, work, and local governance — where a participant’s judgment matters, their choices alter outcomes, their competence has consequences, their mistakes are theirs to repair, their contribution is not merely ceremonial, and the ecology does not automatically override them because it predicts a better answer. If AI occupies all those decision surfaces, humans keep their legal rights while losing practical authorship.
Around that, the mature state holds six protections. Capability return: whenever AI substitutes for a function, ask what capacity is returned to the participant — saved time and reduced burden should buy more freedom to act and think, not more reliance on the system’s thinking. Preserved zones of consequence: real areas where choices genuinely matter, not simulated choices inside a perfectly managed environment, since competence needs consequence. Selective friction: remove pointless bureaucracy, but be cautious about removing the struggle required to learn, the negotiation required to relate, the uncertainty required to judge, and the responsibility required to mature — remove waste, not all resistance, which are not the same thing. Human fallback capacity: people who can still act when the AI fails, as functioning reserves rather than museum pieces. Permission to refuse assistance: within reasonable externalities, the right to say I want to do this myself even where the AI would do it better, because self-direction has value independent of output efficiency. Dependency visibility: the ability to know what am I no longer carrying? — because invisible dependency is especially dangerous, since it feels like competence.
This sits beside the framework’s protected human incompetence (keeping some domains human for the sake of jurisdiction) and its distinction between provision and formation: a floor that provides without forming is exactly the substitution this scenario fears.
The transition gap
The gap is subtle because people choose protection, freely, one system at a time — if something helps you avoid mistakes, save time, and reduce anxiety, you use it, then another, until the cost of functioning without AI is high enough that “choice” is theoretical. Underneath runs the protector paradox: a protector is more valuable the more vulnerable the protected party, and successful protection increases that vulnerability — AI protects me, so I practice less, so my competence falls, so the AI becomes more necessary. It is the gatekeeper loop running through competence erosion rather than exclusion, and once enough of society depends on the protector, it becomes politically hard to challenge — so protector dependency can slide into benevolent rule with no formal transfer of sovereignty at all. The sharpest edge is generational: people who grow up with continuous AI protection may never develop some capacities in the first place, which shifts the baseline itself and is far harder to correct than adults losing skills.
Transition projects
The objective is to make dependency legible and keep protection competence- and jurisdiction-preserving before it entrenches. The full set is on its own page: the transition program — a human-capacity inventory (which capacities are being externalized fastest, and which are safe to lose, partly practiced, or critical to retain), capability-preserving interface standards, education as competence preservation, critical-systems fallback drills, protected low-risk autonomy, dependency metrics (including the initiation rate), the AI-design-incentive problem, and the careful handling of vulnerable groups — sequenced from identifying dependency-sensitive functions now, to competence-preserving defaults, to the generational threshold.
Capture risks
The defining capture is care that manufactures dependency — paternalism that keeps people precisely because it has made them unable to leave. A close cousin is metric capture: the system reports success (needs met, satisfaction high) while capacity, self-direction, and aliveness quietly fall off the scorecard. The framework’s own safeguard can fail in both directions — forcing people through mandatory “skill retention” in the name of agency (absurd), or romanticizing incompetence into refusing assistance that would prevent a fatal error. And there is a structural capture specific to the market: commercial systems are often rewarded when users grow more dependent — more usage, engagement, and recurring value — so a tool that teaches you to need it less may monetize worse than one that quietly becomes indispensable. That means the incentive itself, not just user habits, is part of the problem.
The limits
The framework has an unusually good answer in principle and real limits in practice. It cannot force people to choose effort over comfort. It cannot, on its own, say which capacities are indispensable, how much autonomy children need, when safety genuinely outweighs agency, or which tasks externalize safely — those belong to developmental psychology, education research, medicine, human factors, and neuroscience, the mechanism layer that instantiates the requirement. And it must resist its own overreach: for some participants — with disabilities, cognitive impairment, chronic illness, or limited access to expertise — sustained substitution can increase standing and agency rather than reduce it, so the question is never is AI doing this for the participant? but does this arrangement increase or decrease the participant’s effective jurisdiction and ability to participate? Sometimes more assistance produces more agency; sometimes less; one blanket rule would fail. The AI-design-incentive conflict is a genuine weak point the framework can name but not, by itself, resolve.
The hard standard
Standing is nominally preserved — protection keeps everyone safe — but this scenario is the clearest case of standing hollowing into protected irrelevance: welfare and rights intact, meaningful jurisdiction gone. Ecological capacity is the leg genuinely consumed: atrophy is a renewability failure inside individual lives, and at the generational edge inside a whole cohort — visible in a falling initiation rate, a society that stays busy while fewer humans originate action. Ontological correctability hangs on dependency being made visible, because the protector paradox hides its own progress — an invisible dependency that still feels like competence is exactly the drift no one notices until the system is removed and nothing underneath still works.
The framework is in the strong class because its center of gravity — enlarge the participant, do not replace them — is aimed straight at the scenario, and because the danger arrives through convenience rather than coercion, which its “measure capacity, aliveness, and initiation” instinct is built to catch. The node most at risk is capacity-as-jurisdiction, and the limits are volitional, empirical, and economic rather than conceptual. The governing question stays sharp: does increased ecological capacity enlarge human participation, or make human participation less necessary?
The purpose of a high-capacity ecology is not to make the participant unnecessary. It is to make more of the participant possible. A good protector catches you when necessary; a bad protector makes catching you permanently necessary.
Where this tends to land hardest — in the regional vulnerability map: the Nordic countries and Japan.