Action, for Arendt, is not behavior. Behavior is predictable, patterned, the sort of thing statistics describe and social science models. Action is the unpredictable insertion of a unique being into the web of human relationships through deed and word, and it carries three defining features: it reveals who the actor is (not what she is), it requires plurality — the presence of others who can witness and respond — and it is both unpredictable in its initiation and irreversible in its consequences. In the AI transition, Arendt's framework identifies action as the twenty percent that remains after the machine absorbs the labor and routine fabrication: the judgment, the decision about what deserves to exist, the beginning that could not have been predicted.
Arendt distinguished action sharply from the other two activities in the vita activa. Labor maintains biological life and produces nothing durable; work fabricates the artificial world of objects. Action alone requires other people and alone reveals the actor's identity. The carpenter's chair tells us what the carpenter can do; only the carpenter's speech and deed in the public realm tell us who she is.
The AI moment places enormous pressure on this distinction. A generative model can now produce what looks like action — text that argues, images that persuade, code that initiates consequences. The Arendt simulation insists the resemblance is superficial. Machine output lacks the who: it has no unique biography, no stakes in the consequences, no irreducible perspective that would make the output a revelation rather than a rearrangement.
Action's unpredictability is structurally important for Arendt's political philosophy. Behavior can be governed by rules, optimized by administration, managed by expertise. Action resists all of these because its outcomes cannot be foreseen. This is why Arendt treats action as the locus of freedom: freedom is not the absence of constraint but the capacity to begin, and beginning requires the willingness to act without knowing what the action will unleash.
The Arendt simulation extends this analysis to the AI discourse's characteristic confusion. When commentators celebrate AI's capacity to 'make decisions,' they conflate decision (which reveals a stance) with optimization (which selects within a predefined objective). The judgment economy that The Orange Pill identifies as the new premium is, in Arendtian terms, the economy of action — the premium on genuine decision rather than sophisticated pattern-following.
Arendt developed the concept of action most fully in The Human Condition (1958), building on her earlier work on totalitarianism and drawing on the Greek distinction between praxis (action) and poiēsis (making). The framework received refinement in Between Past and Future (1961) and application in On Revolution (1963). Her mature treatment positioned action as the activity most endangered by modern conditions — a thesis that acquires new force in the AI age.
Reveals who, not what. Action discloses the actor's identity in a way no description of her attributes can capture.
Requires plurality. No action occurs in solitude; the presence of others is constitutive, not accidental.
Unpredictable and irreversible. Action initiates consequences that cannot be foreseen and cannot be undone — features that distinguish it from fabrication.
The twenty percent that remains. When AI absorbs the labor component of professional work, what survives is the space where action must occur — or fail to occur.
The Arendt simulation presses a harder claim than Arendt herself: that AI systems categorically cannot act because they lack natality and stakes. Critics respond that this definition is stipulative and may simply restate the conclusion it purports to reach. Defenders argue that the stipulation is productive: it forces the AI discourse to confront what its casual use of agential vocabulary obscures.