CONCEPT
Care (AI Evaluative Criterion)
The quality of attention brought to AI output evaluation—the willingness to reject the smooth, sit with discomfort, treat collaboration as occasion for judgment rather than extraction—detectable only in the output itself, not measurable by metrics.
Care is the second criterion in the evaluative framework this volume proposes, and it is the criterion that connects formal analysis to ethical practice. Care is the sustained attention brought to the encounter with AI output—the discipline of reading each generated paragraph, examining each line of code, questioning each plausible claim, refusing to accept smoothness as evidence of quality. It is not a feeling but a practice: the practice of treating the AI collaboration not as a productivity tool delivering outputs for consumption but as an occasion for the exercise of judgment. Care is detectable not through self-report (every builder claims to care) but through the output's formal properties: the evidence that outputs were rejected, that iterations occurred, that the final configuration was shaped by discriminating evaluation rather than received by default. Segal's confession of his failures of care—the
Deleuze passage accepted because the surrounding prose was smooth, the moments when Claude's eloquence outran his thinking—is itself an exercise of