Researchers have introduced ProSE, a novel framework for AI assistants that accounts for user bounded rationality when generating proposals. Unlike previous methods that focus solely on proposal quality or user goal inference, ProSE's proposals also serve as probes to learn latent user preferences and evaluation constraints. This approach, formalized as a hidden-parameter sequential assistance problem, uses a KL-regularised bounded-rational binary response model to balance value gain against an evaluability penalty. A depth-2 Bayes-adaptive planner, ProSE-Plan, was developed and demonstrated in simulations to improve over baselines when evaluation cost is a bottleneck, by selecting more informative proposals. AI
IMPACT This research could lead to AI assistants that are more effective by better understanding and adapting to user evaluation limitations.
RANK_REASON Academic paper proposing a new AI assistance framework. [lever_c_demoted from research: ic=1 ai=1.0]
Read on arXiv cs.MA (Multiagent) →
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
