On recognizing the handiwork of AI

Abstract

As AI-generated images and texts proliferate, people have developed techniques for identifying them using clues like misshapen hands in images or distinctive words in text. This commentary situates these emerging practices within what Carlo Ginzburg called the “conjectural paradigm”: a mode of knowing that links contemporary AI detection to older traditions of medical symptomatology, art historical connoisseurship, and detective work. Yet unlike the stable or slowly evolving clues of earlier conjectural practices, the signifiers of AI involvement are rapidly shifting. This instability has consequences not only for how texts are read but also for how they are written. Authors now navigate a landscape of suspicion where their words may be misrecognized as machine generated. Rather than resolving into stable literacies, our efforts to recognize AI’s handiwork reveal the deeper uncertainties of authorship and interpretation.