Skip to content
Entangled SocietyObservatory of Consequences

Delegation / Authority / Accountability

MODEL v0.1 · OBSERVATIONAL FRAMEWORK

FILE HRR-001 / 2026-09-12 / CONCEPT

THE HUMAN
RESERVE RATE

How much work should remain human even when AI can perform it?

Capability does not determine delegation.

Human Reserve Rate

NOT YET QUANTIFIED

UNMEASURED

A numeric rate will not be published until observations accumulate and the method can support one. No placeholder percentage is shown.

Calibration set: 6 observations

Quantification deferred until delegation, authority and accountability can be distinguished reliably across domains.

Read the method.

Central question

If AI can do the work, should AI do the work?

The future of work will not be determined by capability alone.

It will also be determined by what societies choose not to delegate.

01

Definition

What this model observes

This model does not measure what AI can do.

It observes how much of the work AI can already perform a society intentionally keeps under meaningful human control.

Core definition

Human Reserve Rate = AI-capable tasks intentionally retained under meaningful human control / all AI-capable tasks

CONCEPTUAL OBSERVATIONAL MODEL

This is not yet a mathematically precise economic indicator.

It is a conceptual observational model.

No numeric rate is published until the observation set and method can support one.

Human Reserve

Human Reserve is the portion of an AI-capable system in which meaningful human authority is intentionally preserved.

Human Reserve is not the absence of AI. It is the preservation of meaningful human authority.

  • Human Reserve ≠ DENY only

    HOLD, human final authority, intervention rights, meaningful review, and institutional responsibility can also constitute reserve.

  • Human Reserve ≠ Human-only work

    Reserve is not the absence of AI. It is the preservation of meaningful human authority.

  • Human Presence ≠ Human Authority

    A person may remain in the loop while having little influence over what was surfaced, excluded, framed, already executed, or still reversible.

HOLDDENYhuman final authorityhuman intervention rightsmeaningful reviewinstitutional responsibility

Instrument

Delegation Boundary

AI CAPABILITY

AI-CAPABLE TASKS

DELEGATION BOUNDARY

ALLOW

AI EXECUTES

HOLD

SHARED AUTHORITY

DENY

HUMAN AUTHORITY

HUMAN RESERVE

Delegation Boundary

Authority Boundary

Accountability Boundary

02

Delegation

Delegation Model

ALLOW / HOLD / DENY are not a ranking of how dangerous a task is.

They describe whether, and under what condition, an AI-capable task may be delegated.

AI capability is not fed directly into automation. A delegation boundary sits between them.

  1. ↓AI CAPABILITY
  2. ↓TASK BECOMES AI-CAPABLE
  3. ·DELEGATION BOUNDARY

  1. ALLOW

    AI may execute the task within a defined operating boundary.

    AI EXECUTES

  2. HOLD

    AI may assist or propose, but consequential execution remains subject to meaningful human review.

    SHARED AUTHORITY

  3. DENY

    The task remains under human authority even if AI is technically capable of performing it.

    HUMAN AUTHORITY

03

Boundaries

Three Boundaries

  1. 01

    DELEGATION BOUNDARY

    What may be delegated to AI?

    • execution
    • recommendation
    • prioritization
    • diagnosis
    • approval
    • rejection
    • resource allocation
    • escalation

  2. 02

    AUTHORITY BOUNDARY

    Who retains final authority?

    • AI
    • operator
    • professional
    • manager
    • institution
    • regulator
    • distributed responsibility

    Human-in-the-loop and human authority are not the same. A human click is not, by itself, Human Reserve.

  3. 03

    ACCOUNTABILITY BOUNDARY

    Who remains answerable when the decision causes harm?

    • model provider
    • application provider
    • device manufacturer
    • deploying organization
    • professional operator
    • certification body
    • regulator

    Can authority be delegated without accountability also moving?

04

Presence

Five States of Human Presence

H0–H4 describe human presence. They do not rank systems.

These represent different locations of human presence, not maturity.

H4 is not always superior. A laboratory instrument, a clinical authorization, and a reserved civic office require different reserves.

  1. H0 — ABSENT

    AI performs the task autonomously. Human authority is operationally absent.

  2. H1 — OBSERVING

    A human can see what the AI is doing but has little practical intervention capacity.

  3. H2 — REVIEWING

    A human reviews AI output before consequential execution.

  4. H3 — AUTHORIZING

    AI may recommend or prepare action, but a human must explicitly authorize execution.

  5. H4 — RESERVED

    The task remains under human authority even though AI may technically be capable of performing it.

Distinction

HUMAN PRESENCE ≠ HUMAN AUTHORITY

Human Reserve should measure meaningful authority, not mere human presence.

A person may technically remain in the loop while having little meaningful influence over:

  • what information was surfaced

  • what alternatives were excluded

  • how recommendations were framed

  • which path the AI already executed

  • whether reversal remains possible

05

Observations

Observations

Each case observes a narrow AI-capable task and its current delegation state. Product-wide labels are refused. Cases accumulate. They do not yet produce a rate.

Observation

HRR-2026-001 / 2026-09-12

AI Capability Is Expanding Faster Than Delegation Rules

OBSERVED TASK

occupation-to-task economic modeling

Source: Anthropic Institute. Scenarios for our Economic Future

HOLD · H3_AUTHORIZING · Observed + Inference

The classification itself is SHIRO & Co. observational interpretation. It is not an objective scientific fact.

FACT事実

OBSERVED FACT

  • Anthropic Institute published Scenarios for our Economic Future, based on the technical report Economic Scenarios for Transformative AI (Korinek et al., 2026).
  • The model represents jobs as bundles of tasks. AI can leave a task unchanged, augment it, automate it, or lead to new tasks.
  • The published account states that AI capability does not necessarily eliminate an occupation; the internal composition of work changes.
  • The explorer estimates economic futures from expected AI capability, adoption, autonomy, productivity, and labor-market adjustment.
  • The authors state that the model omits policy responses, among other forces, and that the future is not predetermined.

INTERPRETATION解釈

SHIRO & Co. INTERPRETATION

The model primarily estimates what AI may be capable of performing and how firms and workers may adapt economically. It does not fully model the institutional question: which AI-capable tasks should remain under meaningful human authority? This creates a second-order problem. As AI capability expands, societies will need explicit delegation rules determining which tasks may be automated, which require supervision, and which remain reserved for human authority.

Uncertainty

UNCERTAINTY

  • The scenario explorer is Version 1.0 and is described by its authors as a simplification.
  • Omitted forces include policy responses, business cycles, aggregate demand or financial disruptions, and catastrophic risks.
  • This observation does not treat any scenario as a forecast of what will occur.

Signal

The next labor-market variable may not be automation rate, but delegation rate.

Human Reserve implication

Economic models can estimate how much work AI can perform. Institutions will still have to decide how much work AI should be allowed to perform.

Domain
Labor / Economic Systems / AI Governance
Observed capability
Occupations are modeled as bundles of tasks that may remain unchanged, be augmented by AI, become automated, or be newly created.
Delegation state
HOLD
Human presence
H3 — AUTHORIZING
Confidence
Observed + Inference

Classification applies to the institutional gap in the published scenario model, not to Anthropic as a firm.

Delegation boundary

The scenario model estimates how tasks may be augmented or automated. It does not specify which AI-capable tasks may be delegated.

Authority boundary

Economic adaptation by firms and workers is modeled. Final institutional authority over delegation is not assigned.

Accountability boundary

The published model omits policy responses. Answerability for delegated AI-capable work remains outside the scenario frame.

Source links

06

Comparison

Comparison

The same product can occupy more than one delegation state. Each row observes a narrow task, not a vendor.

H0–H4 mark different locations of human presence. They are not a quality ranking and not a maturity ladder.

07

Movement

What changes across the cases

Human Reserve does not disappear all at once.

It moves.

  1. PERCEPTION

    Rokid

  2. JUDGMENT FORMATION

    Halliday

  3. PROFESSIONAL TASK

    Autonomous Diagnosis

  4. SELECTION SPACE

    Hiring Assistant

  5. EXECUTION

    Computer-Using Agent

  6. CONSEQUENCE

    Human Authorization

The relevant question is no longer simply whether a human is present.

The question is where human authority remains in the causal chain.

09

Method

Methodology

MODEL v0.1

Observational framework. Not yet a quantitative economic indicator.

What we observe

We observe situations in which AI becomes capable of performing tasks previously assigned to humans.

What we do not assume

AI capability does not automatically imply permission to automate.

What we classify

Each observation is examined across:

  1. 01

    Delegation

  2. 02

    Human presence

  3. 03

    Authority

  4. 04

    Accountability

What this model is not

  • an employment forecast
  • an AI risk score
  • an anti-automation index
  • a claim that humans should perform every consequential task

Current status

MODEL v0.1

Observational framework. Not yet a quantitative economic indicator.

Calibration set: 6 observations

Quantification deferred until delegation, authority and accountability can be distinguished reliably across domains.

These six observations are calibration cases. They are not averaged, weighted, or converted into a percentage.