AI is better at figuring out what you want by watching your choices than by reading the instructions you actually write for it.
April 1, 2026
Original Paper
Should I State or Should I Show? Aligning AI with Human Preferences
arXiv · 2603.29317
The Takeaway
Most people assume prompt-engineering is the best way to control AI. This experiment reveals humans are so bad at articulating their own preferences that AI predictions based on 'revealed data' are significantly more accurate than those based on a user's own written prompts.
From the abstract
As AI agents become more autonomous, properly aligning their objectives with human preferences becomes increasingly important. We study how effectively an AI agent learns a human principal's preference in choice under risk via stated versus revealed preferences. We conduct an online experiment in which subjects state their preferences through written instructions ("prompts") and reveal them through choices in a series of binary lottery questions ("data"). We find that on average, an AI agent giv