Hot nudges on hazy landscapes

Algorithmic decision-making systems trained on massive data sets, or “assistive AI,” has the potential to remove biases and errors in traditional decision-making, according to some behavioral paternalists. Assistive AI’s democratization of expert advice represents as important and socially beneficial a technological advancement as the internet’s democratization of consensus knowledge. Like any other tool, assistive AI has limitations. I model individuals as theory-based decision-makers whose social systems are open-ended and evolve through time—like “hazy landscapes” with unclear horizons. I consider both traditional nudges and “hot nudges,” automated nudges programmed to learn how to effectively influence their targets by collecting personalized data. I demonstrate that in open-ended social systems “hot nudges” on “hazy landscapes” can exacerbate the knowledge deficits of traditional nudges and may have uniquely pernicious effects on advancing non-beneficial policies and suppressing the emergence of beneficial social institutions.