The Human Agency Framework

What we mean by human agency.

Every governance norm protects or erodes some specific capacity that makes a person an agent. NormSense names the six. Each one is a measurable axis along which AI systems either preserve or diminish what makes us human.

The Foundation

Constitutive ethics

Our framework rests on a tradition running from Kant’s account of rational self-legislation (1785) through Korsgaard’s constitutivist ethics (1996, 2009): an action is morally yours to the extent that you authored it, and authorship requires that you be present — deliberatively, reflexively — in the act of choosing. Agency is not something a person simply has. It is something a person enacts, moment by moment, through the exercise of capacities that constitute them as the author of what follows.

AI systems are powerful precisely because they can intervene at the points where these capacities operate. A norm that affects what a person can see, decide, relate to, be recognized as, know, or choose between is a norm that touches the conditions of human agency itself. This is the moral architecture governance must measure.

From Philosophy to Measurement

Six capacities, each independently measurable

Agency is not one thing. It is a set of distinct capacities, each of which can be eroded or preserved independently of the others. A system can protect what people see while undermining what they can decide. It can grant decisional autonomy while severing the relationships through which choice acquires meaning. The framework names these capacities and treats each as a separate axis of governance evaluation.

A1
Informational
Agency
People cannot act on what they cannot see.
AI systems and their effects are visible to those they affect.
Transparency
A2
Decisional
Agency
Their choices must be their own.
Decisions made by or with AI can be understood and contested.
Fairness
A3
Relational
Agency
Who they become depends on who they can genuinely reach.
Human relationships mediated by AI remain subject to human direction.
Connectedness
A4
Dignitary
Agency
They are people, not resources.
People retain standing as rights-bearing agents in AI-mediated systems.
Empathy
A5
Epistemic
Agency
People cannot direct their path without knowing what shapes it.
The objectives, limitations, and behavior of AI systems are known.
Stewardship
A6
Structural
Agency
Their freedom requires that alternatives exist.
The conditions for genuine choice are preserved as AI systems scale.
Autonomy
Why It Lets Us Measure

From ontology to scoring

Every governance norm touches at least one of these dimensions. A model card requirement is an A1 (Informational) norm. A right to contest automated decisions is A2 (Decisional). A prohibition on synthetic intimacy without disclosure is A3 (Relational). A ban on dehumanizing inferences from biometric data is A4 (Dignitary). A mandate to publish system limitations is A5 (Epistemic). A market-concentration intervention is A6 (Structural).

This commensurability is what makes the platform’s outputs portable. A norm tracked under A2 in healthcare and one tracked under A2 in financial services are comparable not because the regulators agree but because the underlying agency dimension is the same. The framework is the unit of measure.

Read the foundations Request a demo