Sebastian Schmidt on grounding enterprise AI in scientific evidence

Digital Science proposes a shift toward viewing generative models as hypothesis engines rather than standalone truth-tellers. By implementing a semantic layer that utilizes common ontologies and linked data, the company connects proprietary internal research with external publications, patent filings, and clinical trial records. This infrastructure ensures that every chemical formula or experimental parameter is validated against established domain knowledge before it reaches a human expert for final sign-off.

Integrating Scientific Knowledge

For research divisions, the primary challenge remains the isolation of data in internal silos. Connecting these proprietary records with global knowledge networks allows teams to avoid redundant research and uncover hidden development pathways. Instead of relying on noisy internet crawls, the platform uses structured domain logic to filter millions of documents. This allows researchers to pinpoint high-impact prior art and track a scientific concept from initial grant funding through to clinical outcomes. By forcing full end-to-end auditability, the system enables patent attorneys and scientists to trace any automated assertion directly back to its primary evidence, effectively replacing black-box outputs with transparent, verifiable provenance.