| Sumario: | Benchmark-driven approaches to AI ethics increasingly dominate efforts to evaluate artificial systems deployed in socially sensitive domains such as healthcare, education, and criminal justice. By translating complex moral values – like fairness, accountability, or transparency – into quantifiable proxies, benchmarking seeks to deliver clarity, standardization, and comparability. Yet this process risks reducing normativity to measurable outputs, stripping ethics of its contextual, interpretive, and contested character. In response, this paper offers a philosophical intervention grounded in the later work of Ludwig Wittgenstein. Rather than advancing a prescriptive ethical model, it pursues conceptual clarification, challenging the presumption that moral reasoning can be formalized, abstracted, and rendered computationally tractable. From a Wittgensteinian perspective, ethics is not reducible to algorithmic compliance or rule-following, but is best understood as a grammar of practice embedded in language, interaction, and institutional contexts. Benchmarking emerges here not merely as a limited tool, but as a symptom of deeper philosophical misapprehensions – attempts to impose formal closure on normative ambiguity. To move beyond this reductionism, the paper calls for a reorientation toward what it terms "ethical infrastructures": situated, participatory, and reflexive processes that enable moral reflection and accountability across diverse sociotechnical settings. The central question thus shifts from how to measure ethical AI to how ethical meaning is negotiated, sustained, and challenged in practice. Reframed in this way, ethics becomes less about computational alignment and more about sustaining the conditions under which moral reasoning remains plural, situated, and publicly contestable.
|
|---|