Verified, then logged.
Sun 22 Nov 2026 · 00:38 KST / 15:38 UTC한국어 | ENSearch /

Breaking01 / AI

AI proposes excess experiments in mineral exploration test

Researchers reported high recall but low precision when AI selected experiments to test geological hypotheses.

AI-generated editorial illustration for this article

AI proposed more experiments than necessary in a mineral exploration evaluation, researchers reported. They scored experiment selections but did not disclose the number of incorrect selections in the abstract.

The researchers introduced MineralExp, an evaluation tool that asks AI agents to select experiments suited to a geological hypothesis.

The researchers reported high recall, meaning the agents identified a large share of the necessary experiments. Precision—the share of proposed experiments that were necessary—was low.

The team built evaluation cases from published literature on ore deposits. It used a shared classification to accommodate differences in experimental methods for different minerals.

The researchers said domain experts validated the reference experimental plans. A language model also served as a judge when scoring generated plans.

The researchers did not disclose scores for individual evaluated agents in the abstract.