Propose to Learn, Learn to Propose: Evaluability-Aware Assistance under Bounded Rationality

Explain Like I'm Normal

Researchers have developed 'ProSE', a framework that helps AI assistants realize that humans are often too tired or biased to give perfect feedback. Instead of just trying to be right, the AI sends specific proposals to test what the user is actually capable of evaluating. This creates a feedback loop that accounts for human 'bounded rationality' rather than assuming we are perfect logical agents.