Clever Materials: When Models Identify Good Materials for the Wrong Reasons

Kevin Maik Jablonka · 2026

Read Clever Materials: When Models Identify Good Materials for the Wrong Reasons by Kevin Maik Jablonka (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-17730.txt.

Description

Kevin Maik Jablonka tests whether materials models learn chemistry or merely journal clues. Machine learning speeds materials discovery and scores well on many benchmarks, yet strong scores do not prove a model learned chemistry. The paper tests a concrete rival account: a mix-up driven by publication traits.

Across five tasks spanning metal-organic frameworks, perovskite solar cells, batteries and TADF emitters, models trained on standard chemical traits predict author, journal and publication year well above chance. When those predicted byline data serve as the sole input to a second model, scores sometimes rival standard trait-based models. Many data sets thus fail to rule out non-chemical accounts of success, so progress needs disproof tests, tougher data sets, and a clear split between forecast use and evidence of chemical understanding.

Catalog record

CREATORKevin Maik Jablonka
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE29 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-17730

Browse the full library · About the Tower of Babel