The encounter with the face is an event of absolute singularity. One stands before this face, this vulnerability, this demand—and the demand is infinite. There is no limit to what the Other's face asks; no calculus determines that one has done enough. But the world does not consist of one Other. The world is populated by multiple Others—an indefinite plurality of faces, each making its own infinite demand. The arrival of le tiers—the third party, the other Other—introduces a complication that transforms ethics into justice and responsibility into judgment. The third party does not diminish the infinity of the demand. Each face remains infinite. What the third party introduces is the impossibility of responding to one infinite demand without neglecting another. Justice, for Levinas, is not application of universal rules but the painful weighing of infinite demands by a consciousness that cannot satisfy them all.
There is a parallel reading that begins not with the philosophical weight of competing infinities but with the material substrate that makes such weighing possible in the first place. The builder contemplating ethical demands sits atop vast extractive systems—lithium mines in the Democratic Republic of Congo, rare earth processing facilities in China, data centers consuming municipal water supplies, content moderation farms in Kenya where workers label traumatic images for $2 per hour. These are not third parties making claims; they are the invisible foundation enabling the luxury of ethical deliberation.
The Levinasian framework, for all its sophistication about infinite responsibility, operates within a horizon of the already-privileged. The builder who can choose between keeping five team members or achieving the productivity of one hundred has already won a lottery of geography, education, and capital access. Meanwhile, the Congolese child mining cobalt for battery production faces no interesting tension between arithmetic and ethics—only the arithmetic of survival. The real third party problem isn't how to weigh competing infinite demands but how the entire apparatus of AI development depends on making certain faces systematically invisible. The governance frameworks Levinas would find insufficient are not merely incomplete attempts at justice; they are active mechanisms for maintaining the invisibility of the substrate. They create elaborate procedures for weighing the claims of users, developers, and investors while ensuring the claims of miners, moderators, and energy communities never enter the calculation at all.
The builder working with AI is surrounded by third parties—by a plurality of Others whose competing claims cannot all be satisfied. The user who wants a product that works. The worker whose expertise is being rendered redundant by the tool the builder employs. The child who asks what she is for. The community whose cultural norms about work and authorship are being disrupted. The society whose institutions must adapt to changes arriving faster than institutional adaptation permits. Each makes a legitimate claim. Each claim is infinite in the Levinasian sense—irreducible to a number, incalculable by any metric.
Segal's tension between the arithmetic of productivity and the commitment to his team's development is a third-party problem of this kind. The arithmetic said: five can do the work of one hundred. The investor's face demanded efficiency. The team members' faces demanded development. The user's face demanded quality. The displaced worker's face, absent from the room but present in the structural logic, demanded consideration. The child's face demanded a parent present at dinner. Each demand was infinite. The resources were finite. No decision could satisfy all. The decision to keep and grow the team was a decision of justice—not perfect justice, but justice that chose, in the face of competing infinities, to prioritize one set of demands while acknowledging others were not being met.
The AI cannot perform this weighing. Not because it lacks computational sophistication—the machine can optimize, calculate trade-offs, model consequences of different allocations. But the weighing of infinite demands is not an optimization problem. It is an ethical event in which the one who decides is responsible for the decision in a way no algorithm is responsible for its output. The weighing requires the capacity to be claimed, to feel the weight of responsibility, to experience the discomfort of choosing one face over another while knowing the unchosen face does not cease to make its demand.
Contemporary AI governance frameworks—the EU AI Act, national frameworks, corporate governance structures—represent institutional attempts to weigh competing claims of multiple stakeholders. Levinas would acknowledge their necessity while insisting on their insufficiency. They operate in the domain of the Said: explicit, codified, institutionally administered obligation. They produce rules, standards, compliance requirements. But justice in Levinas's framework is never fully captured by the institution. There is always an ethical remainder—a dimension of responsibility no rule covers, no standard anticipates, no enforcement mechanism reaches.
The concept of le tiers is present in Totality and Infinity but developed most fully in Otherwise than Being, where it bridges Levinas's ethical phenomenology with his political philosophy. The third party answers the objection that face-to-face ethics cannot scale: it shows how the plurality of faces necessarily introduces justice, calculation, and institutional structure while preventing these from substituting for the ethical demand that precedes them.
Plurality transforms ethics into justice. The pure face-to-face is unstable once a second face appears.
Infinite demands, finite resources. The tragic structure of moral life is not scarcity of goodwill but the impossibility of meeting all legitimate claims.
Justice with ethical remainder. Institutional weighing of claims is necessary but always leaves an infinity the institution cannot contain.
AI cannot weigh. Optimization is not judgment; the weighing requires a consciousness that can be claimed.
Governance frameworks address the Said. They cannot substitute for the builder's personal encounter with the ethical remainder.
Levinas's move from ethics to justice has been criticized as inadequate to the political problems it invokes. Critics argue that asymmetric responsibility toward individual faces does not scale into workable institutional design. Levinas's political philosophy has been read as underdeveloped compared to his ethics. Defenders respond that the third party's introduction is meant to name the problem of justice, not to solve it—that institutional design requires the ethical remainder's acknowledgment as its ongoing correction. The AI governance debate has given this longstanding question new urgency.
The question of which framework better illuminates AI's ethical challenges depends entirely on the scale at which we examine the problem. At the scale of immediate moral encounter—the builder facing their team, the user encountering bias, the worker being displaced—Levinas's framework captures something essential (90% right). The phenomenology of feeling claimed by competing demands, the impossibility of perfect justice, the irreducibility of ethical remainder to institutional rules—these describe the lived experience of AI's moral complexity with precision that materialist critique cannot match.
At the scale of global systems, however, the contrarian view dominates (80% right). The infrastructure of extraction, the geography of labor arbitrage, the political economy of data accumulation—these aren't competing ethical demands but structured invisibilities. Here, Levinas's framework risks becoming what the contrarian suggests: a sophisticated ethics for those already seated at the table, while the majority who make the table possible remain unseen. The governance frameworks are better understood as capture mechanisms than insufficient attempts at justice.
The synthetic frame emerges when we recognize these as necessarily coupled scales of analysis. The builder's genuine ethical encounter with their team occurs within and depends upon global systems of extraction—this is not hypocrisy but the actual structure of moral life under platform capitalism. The right response isn't to dismiss either scale but to hold both in view: to experience the weight of immediate ethical demands while simultaneously mapping the infrastructures that make such demands possible to consider. The third party, properly understood, includes both the face that appears making its claim and the face systematically prevented from appearing at all.