AI Assistants Should Look Around Before They Act, Researchers Say
Your AI helper may act before it truly understands — and that costs you.
A new paper, "Before Agents Decide: Epistemic Action in LLM-Based Systems," by Yizhi Liu, Balaji Padmanabhan, and Siva Viswanathan, argues that LLM-based agents should consider an earlier question than deciding: is the available evidence ready for the decision? Sometimes necessary evidence is missing; other times evidence is present but its form hides what matters, or the comparison needed to judge it doesn't yet exist. Drawing on cognitive science, the authors call actions that improve the basis for a later choice epistemic actions, and distinguish three modes: acquiring missing evidence, transforming available evidence, and probing a system to create a revealing response. They use the term epistemic scaffolding for the interfaces, tools, and environments that make these actions possible and auditable. The paper argues that agent design must address how decision-ready evidence is produced. Accepted at the FAST Workshop at NeurIPS 2026.
- AI agents (software that acts on your behalf) often decide before they have enough information — the same way a rushed assistant books the wrong flight.
- The paper names three fix-it moves: find missing facts, reorganize messy facts, and test how a system responds before trusting it.
- The authors want AI to show its reasoning steps in ways people can check, so mistakes get caught before money or time is lost.
Why It Matters
Better-behaving AI means fewer wrong bookings, bad purchases, and costly automated mistakes — with less cleanup for you.