AI advances are described as progress, but artificial capability and human capability are not equivalent trajectories.
Research Poster · Human-Development Conversion Constraint
The Human-Development Conversion Constraint: What Should Count as Progress in AI-Rich Societies?
Micheal Charles Preble · Independent Researcher · SSRN 7416900
Problem · Background · Research Question
Builds on the capability approach (Sen 1999; Nussbaum 2000) and extended-cognition scholarship (Clark & Chalmers 1998) rather than claiming either is new.
What additional conditions justify describing technological advancement as human advancement?
Framework: the formal constraint
“A claim that increased artificial capability constitutes human progress is warranted only to the extent that the capability is converted into substantively valuable human possibilities whose distribution and durability are adequate to the scope of the progress claim.”
Central diagram
Stage 1
Artificial capability
What the AI system itself can do.
Stage 2
Practical access
Whether the capability is actually reachable by the relevant population, not merely available in principle.
Stage 3
Effective use
Whether access translates into use that actually functions for the person, given literacy, cost, connectivity, and institutional support.
Stage 4
AI-assisted functioning
The person's functioning while assisted — the level most output-based evaluation stops at.
Stage 5
Retained human capacity
What capacity the person keeps, or builds, independent of the assistance being present.
Stage 6
Substantive human development
Durable, distributed expansion of human possibility — the only stage that would justify a human-progress claim.
Propositions / Observable Implications
- Accounting boundary follows the claim: model-capability evidence supports model-capability claims; human-progress claims require human-condition evidence.
- Distribution is a constituent of progress, not an afterthought — concentrated gains do not establish broad human progress.
- Durability matters: short-term performance gains can conceal path-dependent increases in dependence.
Evidence Required
Illustrative, not original: Bastani et al. (2025) and Strömberg et al. (2026), the same two studies used in the companion paper Beyond Performance, showing assisted gains alongside unassisted declines.
Limitations
Normative, not causal. Does not supply a single metric. The paper explicitly engages — without fully resolving — objections that the constraint may be empirically premature and that human-development metrics can be gamed.
Falsifiers / Open Questions
Not formally falsifiable in the way an empirical hypothesis is; the paper instead states the conditions under which its own inference would be unwarranted (Section 10). Whether causal attribution between AI adoption and human-development outcomes is achievable in specific domains remains open.
Citation / QR
Suggested citation
Preble, Micheal Charles. “The Human-Development Conversion Constraint: What Should Count as Progress in AI-Rich Societies?” Available at SSRN 7416900, 2026.
Preble, Micheal Charles. “The Human-Development Conversion Constraint: What Should Count as Progress in AI-Rich Societies?” Available at SSRN 7416900, 2026.
research.perfinitive.com/publications/human-development-conversion-constraint