Cognitive AI beyond prediction: toward reasoning and discovery

Abstract

Artificial intelligence (AI) has become indispensable in the investigation of materials and systems like batteries, facilitating expedited evaluation of electrolytes, interfaces, and structural frameworks. While modern AI excels in property prediction, it is constrained in its ability to achieve fundamental scientific objectives, including understanding, explanation, and adaptive reasoning. This opinion proposes that AI advancements in materials discovery are developing toward scientific reasoning, encompassing the ability to construct representations, infer mechanisms, formulate hypotheses, conduct experiments, and revise beliefs. A modular cognitive architecture, informed by recent AI developments, can integrate these capabilities to address emerging issues in battery research, such as interfacial instability, failure analysis, and electrolyte design under uncertainty. This paradigm draws inspiration from recent advances in neuro-symbolic reasoning, hypothesis generation, and autonomous systems, which provide the conceptual and technical foundations for future implementation. Cognitively enabled AI systems can work as reasoning colleagues for scientists, enhancing discoveries while maintaining the epistemic integrity of scientific activity.