Asymmetric Understanding (Harris's Analysis)

In the attention economy, platforms accumulated behavioral data—clicks, scrolls, dwell times—to construct shadow models of user preferences. The models were powerful but indirect, inferring mental states from behavioral traces. In the AI economy, the asymmetry intensifies: users provide cognitive data directly through natural language prompts, describing what they think, what they want, what they're uncertain about. The AI processes this data and responds in ways calibrated to the user's revealed cognitive state. The user, meanwhile, cannot inspect the system's internal representations, cannot audit its training data, and cannot evaluate whether its responses serve the user's genuine interests or merely the engagement metrics the system was optimized for. This asymmetry—the system understands the user far better than the user understands the system—creates a power imbalance analogous to the one that characterized the attention economy, but operating at the level of cognition rather than behavior. Harris argues this asymmetry is the central governance challenge of AI, because every regulatory intervention depends on users