AI and Human Capacity
Artificial intelligence can develop, enable, preserve, restore and extend human capacity. It also changes where operative capability resides and what performance legitimately demonstrates about a person.
Expansion and enablement
AI can make knowledge, analysis, communication and creative tools available to people who could not otherwise obtain them. It can reduce unnecessary cognitive and administrative demands, translate information into usable forms, support learning and planning, and compensate for temporary or lasting limitations.
Assistance is not inherently inferior to unaided performance. A system may enable meaningful participation, restore access or extend what a person can accomplish. The relevant question is how a particular configuration changes capability and agency for this person, task and context.
Where capability resides
When a person and AI act together, observed performance may depend on capability retained by the person, capability supplied by the system, capability distributed across both or capability that emerges only through their interaction. These locations can change as the task, model, interface or person changes.
This does not require treating one location as universally preferable. It requires accurate attribution. Institutions need to know whether they are evaluating the person, the tool or the configuration, especially when decisions concern learning, qualification, responsibility or access.
Development over time
AI mediation can support development by explaining, modelling, prompting and providing feedback. It can preserve capacity when a person is overloaded, accommodate limitations and enable recovery or renewed participation. It can also substitute for practice, narrow opportunities to exercise judgment or create dependencies that become visible when access is withdrawn.
These effects cannot be inferred from a single successful performance. They require observation across time: what the person learns, retains and can transfer; what remains possible without the system; and whether reliance is chosen, necessary, reversible and aligned with the person’s purposes.
Performance and bounded inference
An AI-mediated result does not by itself establish what the person could produce independently. Nor does it establish incapacity: the configuration may be the appropriate form through which the person participates. Conclusions must remain bounded by the evidence and by the purpose of the assessment.
Configuration validity asks whether the particular person-system arrangement is appropriate for the task, context and consequences. A configuration suitable for access or production may not answer a question about retained knowledge, just as an unaided test may be invalid when it excludes a legitimate accommodation.
Representations of people
AI also mediates institutions by constructing profiles, classifications, predictions and summaries about people. These representations can influence access, evaluation and treatment even when the represented person never used the system. The capacity question therefore extends beyond user interaction.
Representational agency concerns whether people can know that a consequential representation exists, understand its role, correct errors, contest interpretations and obtain remedy. Governance must address evidence, responsibility and institutional use, not only model performance.
Research and governance
Responsible deployment requires clarity about the capability being relied upon, the evidence supporting attribution, the validity of the configuration, who remains answerable and when reassessment is required. It also requires attention to distribution: who gains new capability, who bears new demands and whose future options are altered.
This remains an active research area. The distinctions presented here organize questions for measurement and institutional design; they are not a finalized instrument, accredited standard or legal authority.