Intelligence as Agency: Evaluating the Capacity of Generative AI to Empower or Constrain Human Action

The advent of generative artificial intelligence (GenAI) has ushered in an era of more open-ended computer systems. Trained to learn patterns from vast datasets of human expression, these technologies exhibit tremendous flexibility in the kinds of input they can parse and produce output that is far more contextually responsive to open-ended interactions with users. Faced with these rapidly burgeoning capabilities, researchers and designers are confronted with an urgent question: how should GenAI be harnessed in ways that empower rather than disempower people? We argue that addressing this question requires three fundamental conceptual shifts to redefine: (1) intelligence as agency, the capacity to meaningfully act, rather than the capacity to perform a task; (2) design as the delegation of constrained agency, rather than the specification of affordances; and (3) ethics as care, an ongoing relation of stewardship toward AI agents, rather than one-time responsibility audits. Developed through an interdisciplinary dialogue between computer science and anthropology, this framework centers convergent interests in design as the intentional shaping of cultural patterns. It offers technologists not only a novel descriptive instrument, capable of better characterizing the outputs and effects of GenAI models, but also suggests possible evaluative methods emphasizing the cultural bases of agentive intelligence.