AI Is Only as Trustworthy as the Judgment of the Person Verifying It

Aug 20, 2026ยท
Woonghee Lee
ยท 1 min read
Abstract
An invited essay on how agentic AI shifts the researcher’s role, told through three scenes from ongoing work: recovering numerical data from published figures, drafting and verifying manuscripts, and consolidating scattered research records. It argues that good design lies not in delegating more to AI but in deciding where it should stop, and that hallucination is prevented not by a smarter model but by a structure that can say “I don’t know.”
Type
Publication
The Window of KAST, Summer 2026 ยท Policy magazine of the Korean Academy of Science and Technology
publications

The Window of KAST, Summer 2026 โ€” policy magazine of the Korean Academy of Science and Technology (KAST). Part of the annual series Research in the Age of AI, under the theme “The Coevolution of AI and Science.”

The article is written in Korean.

The essay follows three scenes from ongoing research โ€” recovering numerical data from published figures, drafting and verifying manuscripts, and consolidating scattered research records โ€” to trace how the researcher’s role moves as agentic AI enters the lab.

Two claims anchor it: good design lies not in delegating more work to AI but in deciding where it should stop, and hallucination is prevented not by a smarter model but by a structure that can say “I don’t know.”

“What we must build is not a smarter model but a more trustworthy system โ€” and that system is only as trustworthy as the judgment of the person verifying it.”

Roughly 2,000 copies were distributed to KAST members, the National Assembly, government ministries, press, universities, and academic societies.

AI Agents Research Automation Trustworthiness Reproducibility Science Policy