A Human Audit of AI-Generated Theoretical Chemistry: Lessons from Ten Chapter-Level Peer Reviews
- 1Polish–Japanese Academy of Information Technology
- 2Chair of Theoretical Physics of Naturally Intelligent Systems (NIS
- 3European Union
Abstract
AI-assisted study spanning molecular stability, the Born–Oppenheimer approximation, chemical bonding, density functional theory, electron correlation, reaction dynamics, thermodynamics, kinetics, and transport. Across the available referee reports, all ten chapters were recommended for acceptance after minor revision within the assessment rubric used in the review project, and no confirmed substantive error was identified that invalidated a principal scientific conclusion. The main value of the project does not lie in claiming ten new fundamental theorems. Rather, the chapters repeatedly make explicit the mathematical steps separating one level of description from another: stability from binding, exact theory from approximation, finite-system calculation from an asymptotic limit, microscopic dynamics from effective kinetics, and conservation laws from macroscopic transport. The review record therefore suggests that AI can assist in constructing technically sophisticated and auditable theoretical arguments. It also motivates, but does not by itself establish, a broader hypothesis: as scientific generation becomes easier, independent verification may become an increasingly important bottleneck.
Related
- Non-anomalous axions: Lessons from the Majoron
- Theoretical description of the energy flow mechanisms in photoexcited semiconductors
- On the description of chemical bonding using the electrostatic potential. I. Theoretical introduction and covalent interactions
- Disorder-tuned soliton nucleation for vibrational energy harvesting in discrete nonlinear lattices. I. Theoretical framework for metastructure design
- Towards a unified view of growth and remodeling models of soft tissues: a comparative review of theoretical frameworks
BibTeX
@misc{sienicki2026human,
title = {{A Human Audit of AI-Generated Theoretical Chemistry: Lessons from Ten Chapter-Level Peer Reviews}},
author = {Mikołaj Sienicki and Krzysztof Sienicki},
howpublished = {ChemRxiv},
year = {2026},
month = sep,
doi = {10.26434/chemrxiv.15008284/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15008284/v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - A Human Audit of AI-Generated Theoretical Chemistry: Lessons from Ten Chapter-Level Peer Reviews
AU - Mikołaj Sienicki
AU - Krzysztof Sienicki
PY - 2026
DA - 2026/09/03
PB - ChemRxiv
DO - 10.26434/chemrxiv.15008284/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15008284/v1
AB - AI-assisted study spanning molecular stability, the Born–Oppenheimer approximation, chemical bonding, density functional theory, electron correlation, reaction dynamics, thermodynamics, kinetics, and transport. Across the available referee reports, all ten chapters were recommended for acceptance after minor revision within the assessment rubric used in the review project, and no confirmed substantive error was identified that invalidated a principal scientific conclusion. The main value of the project does not lie in claiming ten new fundamental theorems. Rather, the chapters repeatedly make explicit the mathematical steps separating one level of description from another: stability from binding, exact theory from approximation, finite-system calculation from an asymptotic limit, microscopic dynamics from effective kinetics, and conservation laws from macroscopic transport. The review record therefore suggests that AI can assist in constructing technically sophisticated and auditable theoretical arguments. It also motivates, but does not by itself establish, a broader hypothesis: as scientific generation becomes easier, independent verification may become an increasingly important bottleneck.
ER -