Artificial intelligence tools expand scientists’ impact but contract science’s focus
Nature • • world
Key Points:
- A fine-tuned BERT-based language model was developed to identify AI usage in research papers through a two-stage training process with specifically constructed positive and negative data sets.
- Expert evaluation of 1,320 randomly sampled papers showed strong agreement among experts and high accuracy of the language model's AI usage identification results.
- AI-related papers consistently receive significantly more citations than non-AI papers across different eras, indicating a higher academic impact.
- Researchers adopting AI publish on average 3.02 times more papers annually than their non-AI counterparts across six scientific disciplines, demonstrating greater productivity.
- AI adoption is linked to smaller research teams, primarily due to fewer junior scientists, and accelerates the career progression of researchers from junior to established roles, while also
