Proposed research direction · Version 0.1

Human-Centric Economic Impact

A measurement layer for asking what changed in the human when AI produced an economic gain.

Review the proposed measuresReview the surrounding field
No results

The same productivity gain can conceal different human outcomes.

Two people may each double their output with AI. One may develop capability that transfers to later work. Another may lose performance when the tool, model, or accumulated relationship state disappears. A third may continue to perform well because valuable capability resides in a governed human–AI relationship.

Conventional output measures can record the gain without distinguishing these conditions. This proposal asks how economic measurement changes when the trajectory, agency, and resilience of the human participant are treated as part of the result.

Position in the field

An additional layer, not a competing observatory

The Stanford Digital Economy Lab’s AI Economic Indicators organizes timely evidence around employment and wages, aggregate outcomes, consumer surplus, hiring and skills, and AI usage. That work establishes the economic observation problem this proposal enters.

The added question concerns conversion. When measured output rises, how much of that change becomes durable human capability, how much remains supplied by the system, how much depends on a continuing relationship, and how is the resulting value distributed?

YHHuman baseline

Performance the person can demonstrate before the measured period of assistance.

YH+AIAssisted output

Performance produced by the active human–AI configuration.

YH′Later human state

Performance, judgment, agency, and opportunity observed after assistance or under changed conditions.

Research lineage

The economic proposal is an expression of the larger corpus.

The earlier papers establish what an aggregate measure would otherwise compress. The Operant Dyad, The Persistent Primitive, and Emergent Novelty locate consequential capability across the person, the system, and the relationship formed through repeated interaction.

Authority Does Not Travel by Default and Governing Relationship State show why control over accumulated state belongs inside the economic analysis. Portability, dependency, and value cannot be evaluated fully without asking whether later reliance remains visible, contestable, and authorized.

The Human-Development Conversion Constraint and Beyond Performance supply the evidentiary boundary. A gain observed in the assisted configuration does not establish a durable human gain until the relevant functioning, distribution, authority, and time horizon have been specified and examined.

This proposal combines those structures into a measurement question. It remains an early research design, not an empirical conclusion.

See how the full corpus fits together

Proposed measurement layer

Six distinctions beneath the aggregate result

These are candidate dimensions for study. They are not a validated index, and version 0.1 does not assign weights or combine them into a welfare score.

Assisted performance

Change in speed, quality, error, completion, or output while the AI system is present.

Capability conversion

Improvement that remains observable when assistance is removed or the person encounters a meaningfully changed task.

Human agency

The person’s practical ability to inspect, correct, refuse, supersede, and redirect the system’s contribution.

Dependency and fragility

Performance lost when a tool is withdrawn, a model changes, or accumulated relationship state is unavailable.

Governed extension

Capability located in a continuing human–AI configuration whose state remains visible, contestable, and under human judgment.

Value capture and distribution

How gains and burdens appear in wages, time, mobility, job quality, ownership, bargaining power, and exposure to loss.

Candidate diagnostic

Human Capability Conversion Ratio

A first diagnostic could compare the human improvement observable after assistance with the larger gain observed while assistance was active.

HCCR = (YH′ − YH) ÷ (YH+AI − YH)Candidate measure. The terms require task-specific operational definitions.

A ratio near one would suggest that much of the observed assisted gain remained detectable in later human performance. A value near zero would suggest that the gain remained primarily configuration-dependent. Negative values could indicate deterioration, while values above one could reflect continued learning, measurement noise, or a task that changed the underlying comparison.

The ratio is undefined when assisted performance does not exceed baseline. It cannot establish human development by itself, and comparisons across unlike tasks may be invalid. Agency, opportunity, distribution, and governance remain separate parts of the proposed economic account.

Proposed first study

A longitudinal conversion design

The first study should isolate one bounded task family and make the transition between assisted and later performance visible.

Baseline

Observe unaided performance, confidence, error recognition, and task strategy before the measured assistance period.

Assisted work

Record configuration-level output together with corrections, overrides, time, and the system contribution retained in relationship state.

Transfer

Test later unaided performance and a related novel task without assuming that speed alone represents learning.

Transition

Change the model or remove accumulated state to observe dependency, portability, resilience, and the role of governed continuity.

Study status: This sequence is a design proposal. A pilot requires preregistered hypotheses, validated task measures, an explicit consent and data-governance plan, and independent review before results are interpreted as evidence of human development.

Research foundations

Existing measures constrain the proposal

The International Labour Organization treats AI-exposure indicators as signals of possible task change rather than forecasts of employment outcomes. Its decent-work framework also keeps freedom, equity, security, dignity, and job quality within view.

The OECD Well-being Framework asks whether economic change improves life and for whom. The NIST AI Use Taxonomy supplies human-centered, technique-independent language for classifying how AI contributes to a task and its intended outcome.

This proposal does not claim that human-centered economic measurement is absent. Its narrower contribution is to connect those concerns to a longitudinal conversion test and to distinguish individually retained capability from capability sustained through a governed relationship.

Limits and open questions

What version 0.1 does not establish

The framework does not determine which outcomes people should value, reduce human development to task performance, or establish a causal effect of AI from observational data. It does not assume that unaided performance is always superior to governed extension, or that all dependency is harmful.

Substantive questions remain about task selection, baseline instability, learning and decay, disability and accommodation, organizational power, cultural variation, employer surveillance, and who has authority to define an acceptable human outcome. Those questions affect the construct, not merely its implementation.

Version record

Public revision history

  • v0.1September 5, 2026 — initial research direction: field position, six measurement dimensions, candidate conversion ratio, proposed longitudinal sequence, and limits.

Material revisions should preserve the prior version, identify what changed, and explain how the revision improves or narrows the construct.

Return to the research programMethods and disclosures