AI Now Reads Medical Research Faster Than Humans
This could save scientists months of work — and speed up pandemic responses
Deep Dive
Scientists tested large language models to extract information from 536 peer-reviewed agent-based modeling papers on disease spread, comparing results to a human-conducted review. The models got paper-level accuracy around 78–82%, but field-level accuracy ranged from about 32% to 100%, with more complex or subjective fields proving less reliable. Notably, agreement between models was a potential quality signal: low agreement may suggest hallucinations, while high agreement paired with low accuracy could point to errors in the human dataset.
Key Points
- AI can scan 536 medical papers about disease spread with ~80% accuracy — faster than humans
- When AI agrees with itself, it’s more likely correct; when it disagrees, it might be making things up
- This could cut months off research time, helping scientists respond faster to health crises like pandemics
Why It Matters
AI could help scientists work faster, potentially saving lives by speeding up medical discoveries and crisis responses.