Capabilities vs. Functionings
A functioning is something that exists in the world — an essay written, code compiled, a product shipped. A capability is what a person can actually do and be — the internal capacity developed through the process of producing. The distinction matters because justice requires providing capabilities, not mandating functionings; and because, under AI conditions, the relationship between the two has become severable in historically unprecedented ways. A world full of extraordinary outputs can also be a world in which the people producing them are becoming less capable. The product improves while the person plateaus. The functioning exists. The capability does not.
In the AI Story
The distinction is foundational to the capabilities approach Nussbaum developed with Amartya Sen. The approach evaluates societies not by their aggregate production but by what real freedoms — what genuine opportunities for functioning — are available to each individual. This refusal to measure in aggregates alone is what makes the framework uniquely suited to the AI transition, where aggregate productivity metrics can be rising while individual capabilities are quietly being hollowed out.
Applied to AI-assisted work, the distinction generates questions that productivity metrics cannot ask. The student who uses AI to generate an essay has produced a functioning — the essay exists. But she may not have developed the capability for sustained argument that writing the essay was designed to cultivate. The engineer who uses AI to bypass debugging has shipped working code. But she may not have developed the architectural intuition that debugging would have deposited. The functioning is real; the capability it should have generated may be absent.
The point is not hostile to AI. It is a refusal of the equation of output with value. Some AI uses expand capabilities — the engineer who delegates mechanical implementation to focus on architectural judgment may genuinely deepen her capacity for thought. Other uses contract them — the user who treats the tool as a replacement for cognitive struggle acquires functionings while atrophying the capacities those functionings are supposed to express.
The 2025 paper by Ratti and Graves — A Capability Approach to AI Ethics — argues that through this lens, AI ethics becomes the investigation of AI's impact on Nussbaum's central capabilities, not on outputs or productivity. The question is not whether AI produces more but whether the conditions under which AI-assisted work occurs permit the full development of human capabilities.
Origin
The distinction was developed by Nussbaum and Amartya Sen in the 1980s and 1990s, initially as a critique of welfare economics that measured well-being through utility or resource possession. Sen's Development as Freedom (1999) and Nussbaum's Women and Human Development (2000) established the framework's core architecture, which Nussbaum elaborated in Creating Capabilities (2011).
The framework was adopted by the United Nations Development Programme as the basis for the Human Development Index and has shaped development policy in dozens of countries. Its application to the AI transition is more recent — emerging through the 2023–2026 work of philosophers including Ratti, Graves, and researchers applying the framework to automation and displacement.
Key Ideas
The severability problem. AI has made it possible, for the first time in the history of human tool use, to produce functionings without developing the capabilities those functionings normally express.
Invisibility of the loss. Because the output looks the same, the atrophy of capability is undetectable by any metric that measures only what is produced.
Real freedom, not menu selection. A capability is the substantive opportunity to exercise a human capacity — not the ability to select from options someone else has generated.
The evaluative question. The relevant question about any AI deployment is not what was produced but what the human was able to do and be in the process of producing it.
Institutional implications. Just institutions protect not outputs but the conditions under which capabilities are developed and exercised.
Appears in the Orange Pill Cycle
Further reading
- Martha Nussbaum, Creating Capabilities: The Human Development Approach (Harvard University Press, 2011)
- Amartya Sen, Development as Freedom (Knopf, 1999)
- Martha Nussbaum, Women and Human Development: The Capabilities Approach (Cambridge University Press, 2000)
- Emanuele Ratti and Mark Graves, 'A Capability Approach to AI Ethics,' American Philosophical Quarterly (2025)