SOT 2026

March 22, 2026

Evidence Prime took part in the SOT 65th Annual Meeting and ToxExpo in San Diego on 22–25 March 2026. Artur Nowak contributed to the continuing education programme, sharing a technical perspective on how AI-assisted evidence work can remain transparent and open to expert review.

Practical AI for hazard and risk assessment

On Sunday, 22 March, Artur presented as part of course PM10, This Is Real: How AI Can Actually Be Used in Hazard and Risk Assessments. The afternoon course connected current applications of AI with the documentation, validation and professional judgment needed in risk assessment.

Andrew Rooney and Seneca Fitch chaired the session. Contributions covered several parts of the workflow.

The continuing education course on practical AI in hazard and risk assessment at SOT 2026.
The continuing education course on practical AI in hazard and risk assessment at SOT 2026.

How the parts of an extraction workflow fit together

Artur’s presentation material examined how specialised agents can combine deterministic rules, traditional machine-learning methods and large language models. Examples included unit conversion, parsing table cells, populating fields only when they apply, and checking extracted results.

The architecture described a controller that coordinates these tasks and records the tools, inputs and outputs involved. It also considered how shared vocabularies and project-specific dictionaries can help humans and AI use consistent terms. Human oversight was part of the design, including the review of uncertain outputs and suggestions that do not fit an existing vocabulary.

These details matter because evidence extraction produces information that other people will analyse and use. Reviewers need to understand where a value came from, how it was transformed and which decisions require their judgment. A useful workflow must make that chain available for inspection.

Keeping professional judgment visible

The course brought developers, regulators and researchers into the same discussion about practical AI use. For Laser AI, it reinforced the importance of evaluating a complete evidence workflow, documenting its limits and preserving clear points for expert review. Artur’s conference recap shares more from the session.

Blank white image with no visible content or details