it would be interesting if an AI with semantic understanding could read and understand all the material and you could interrogate it about the backup behind a claim.

it would be interesting if the entire lab notebook record was available to such an AI and cryptographically signed and dated. that would potentially prevent people from doctoring images and data, as has been recently alleged on multiple occasions.

it would be interesting if that AI was trained to do causal inference and do-calculus, and every paper published generated an incremental update to an automatically constructed meta-analysis and systematic research review.

keep going and make a large body of real-world e.g. medical data available to the AI, and every interaction could be an addition to the clinical trial data. and then make the AI able to generate new hypotheses and directions for experimentation to test those hypotheses.