Toward a Theory of Critical AI Literacies as Media Literacies
TL;DR
A teachable framework for critical AI literacies is proposed that shares key outcomes with frameworks for critical media literacies: the ability to competently navigate interface persuasion, recognize the incentives found in product design, identify simulations of empathy and care, and make safer decisions about when—and whether—to treat chatbots as confidants.
Abstract
Personified chatbot interfaces have become the public's primary encounter with large language models, encouraging users to interpret their fluent outputs as if they delivered embodied and humanlike care, understanding, and presence. This think piece argues that the most consequential risks posed by contemporary chatbots result from a widening gap between what these highly engineered technologies are and what their most avid users experience them to be. Building on the well-known phenomenon of the ELIZA effect, the article shows how design choices—including bots that use first-person voice, affect-laden language, companion branding, and engagement-maximizing interaction—create a confessional interface that enables affective disavowal, a stance in which users can simultaneously know a bot is not human and still relate to it as if it were. These dynamics intersect with documented harms, including dependency, reinforcement of delusional spirals, and unsound guidance. Rather than treating such outcomes as anomalies attributable to “user error,” the article frames them as predictable effects of sociotechnical systems shaped by attention-driven markets and anthropomorphized interface rhetoric. In response, it proposes a teachable framework for critical AI literacies that shares key outcomes with frameworks for critical media literacies: the ability to competently navigate interface persuasion, recognize the incentives found in product design, identify simulations of empathy and care, and make safer decisions about when—and whether—to treat chatbots as confidants.