Reviewing Artificial Intelligence in Materials Discovery

Materials science is a branch of science and engineering that studies materials and attempts to understand their properties in an effort to improve their performance.¹ This field has led the exploration of building new and improved technologies such as batteries and polymers. In addition, new advancements in spectroscopy and data analysis have helped fuel this growth.

The most recent developments were discussed in a peer-reviewed article published in Next Materials.² The review article focused mostly on how machine learning (ML) and deep learning (DL) models are being used to analyze trace evidence such as fibers, paints, glass, soils, polymers, metals, and chemical residues.¹ The conclusion is that artificial intelligence (AI) is already having an impact on improving the objectivity of forensic evidence analysis.¹

How is artificial intelligence (AI) being used in materials science?

AI has been used to enhance many spectroscopic tools that are commonly used in forensics and forensic analysis. These tools include Fourier-transform infrared spectroscopy (FT-IR), Raman and surface-enhanced Raman spectroscopy (SERS), gas and liquid chromatography–mass spectrometry (GC–MS/LC–MS), and hyperspectral imaging.¹ All these tools have already demonstrated measurable gains in discrimination power, throughput, and objectivity in operational laboratories.¹

Why are the current AI enhancements of forensic tools important?

One of the biggest challenges forensic scientists, legal professionals, and laboratory administrators face is that current evidence-screening workflows mostly only allow for limited throughput, subjective interpretation by analysts, and the difficulty of working with small, degraded samples.¹ By adapting AI frameworks already validated in materials discovery, the authors argue forensic laboratories could process more evidence with less analyst-to-analyst variability.¹

What gains have scientists made in AI-assisted trace evidence classification?

In the review article, the research team cited several key examples where AI is helping scientists make gains in trace evidence analysis. As one example, FT-IR and Raman spectroscopy have utilized AI to classify fibers, drugs, and paints.¹ Researchers also cited probabilistic likelihood-ratio reporting for glass fragments and gunshot residue and AI-enhanced hyperspectral imaging for document authentication as other additional advancements.¹ The review identifies these as technologies approaching operational maturity between 2025 and 2030.¹

What are some of the legal and procedural constraints researchers highlighted in their review article?

The review places heavy emphasis on legal and procedural constraints that distinguish forensic AI from its industrial counterparts. Any system used in casework must satisfy quality frameworks such as ISO/IEC 17025 accreditation and meet U.S. Daubert admissibility standards for scientific evidence, which demand explainability, rigorous validation, and reproducibility.^1,3 The authors argued in their paper that these requirements must shape system design "from the ground up" rather than being retrofitted after development.¹

What do the researchers anticipate will come in the future?

The research team highlighted in their article that uncertainty-aware ML methods, such as conformal prediction, Bayesian neural networks, and calibrated ensemble models, will become more widely used. These methods, according to the researchers, have key advantages. For example, uncertainty-aware ML methods generate probabilistic outputs in the form of likelihood ratios rather than simple binary matches, which is a scientifically defensible and legally necessary format.¹ Other current priorities that the researchers see coming to fruition include combining spectroscopic data, imaging, and case-provenance metadata into unified analytical frameworks, and transferring "foundation models" trained on large materials-informatics databases to forensic-specific tasks as a way to work around limited forensic training data.¹

Meanwhile, in the long term, the research team believes that federated learning networks will allow AI models to be trained across laboratory databases in different jurisdictions without centralizing sensitive case data.¹ This structure, according to the authors, will ultimately be needed to achieve the statistical robustness required in criminal proceedings.¹

The authors conclude that, if rigorously validated and ethically deployed, AI could enhance not only laboratory efficiency, but the broader reliability and fairness of the criminal justice systems that rely on forensic evidence.¹

References

1. Colorado State University, School of Material Science & Engineering. Colorado State University, 2026. https://www.research.colostate.edu/smse/ (accessed August 25, 2026).
2. Datta, S.; Chakraborty, R.; Bhattacharjee, R.; Kar, A. K. A Review of Artificial Intelligence in Materials Screening: Data, Models, and Applications in Forensic Science. Next Mater. 2026, 13, 102756. DOI: 10.1016/j.nxmate.2026.102756
3. International Standards Organization (ISO), ISO/IEC 17025. ISO, 2026. https://www.iso.org/ISO-IEC-17025-testing-and-calibration-laboratories.html (accessed August 25, 2026).