AI for scientific discovery: How researchers use Claude (www.anthropic.com)

Researchers have conducted an extensive analysis of how AI model Claude is utilized in scientific discovery, covering over 207,000 sessions to assess its impact across various fields. They found that Claude is predominantly used in formal sciences like computer science, mathematics, and statistics for tasks such as data analysis and writing, while its application in health and lab-based physical sciences is less frequent. Notably, data analysis sessions account for one-third of Claude's usage, highlighting where AI capabilities are embraced most effectively. The study introduced a measure of AI's theoretical capability to perform research tasks, revealing that tasks where AI excels see up to four times higher usage than predicted, indicating a trend towards optimizing AI’s strengths in cognitive tasks. The implications of this research point to a potentially transformative impact of AI on research output. If current AI capabilities are widely adopted, there’s potential to double research productivity in certain domains, with projections suggesting a 15-fold increase in fields heavily reliant on non-physical research tasks, such as computer science. However, the study also flags potential risks, including ethical concerns and shifts in researchers’ career dynamics, as AI takes on more responsibilities in the research landscape. This highlights both the opportunity for accelerated discoveries and the importance of ensuring responsible usage of AI in scientific inquiry.