Recent advances in the SISSO method and their implementation in the SISSO++ code

Explainability as a core feature: The paper demonstrates that symbolic regression methods like SISSO create "relatively simple analytical descriptions" of material properties, making AI models interpretable to scientists and stakeholders—a key requirement for responsible AI in high-stakes domains like materials discovery.

Deterministic methodology reduces bias: SISSO's deterministic nature (as opposed to stochastic methods) provides reproducible and auditable results, which is crucial for responsible AI applications where decision-making processes need to be transparent and verifiable.

Expanded capability through controlled optimization: The introduction of "controlled non-linear optimization" to feature creation broadens the types of problems SISSO can solve while maintaining explainability, suggesting that responsible AI doesn't require sacrificing model capability for transparency.

Improved reliability and efficiency: The paper reports "drastically increased reliability and efficiency" in the solver algorithms, which directly supports responsible AI deployment by reducing computational waste and improving the consistency of results across different runs.

Practical implementation focus: By providing full mathematical and computational details and releasing SISSO++ as an open-source tool with Python bindings, the authors enable broader scientific adoption and scrutiny of the method, supporting the responsible AI principle of accessibility and reproducibility.