Ubiquity of AI Use in Scientific Practice

Determine whether and to what extent the use of artificial intelligence as part of scientific practice is becoming ubiquitous within and across scientific and scholastic fields.

Background

The study measures AI engagement across roughly 80 million publications using a binary keyword-based classifier that does not distinguish between different forms of engagement (e.g., practical use of AI tools in scientific workflows, development of new AI methods, or conceptual/critical discourse about AI). While the authors demonstrate rapid growth and diffusion of AI-engaged publications, their methodology cannot identify whether this engagement reflects actual use of AI in scientific practice.

Because the analysis aggregates all forms of AI engagement, the authors explicitly note that it remains unclear whether the practical use of AI in conducting research is itself becoming ubiquitous. Resolving this uncertainty requires approaches that differentiate types of AI engagement and directly assess the extent of AI tool use across fields.

References

Given that our study relies on a binary classification technique, we are limited in what we can say about the nature of AI engagement. For instance, it is not clear from our work whether and to what extent the use of AI as part of the practice of science is increasing in ubiquity.

— Oil & Water? Diffusion of AI Within and Across Scientific Fields  (2405.15828 - Duede et al., 2024) in Discussion and Concluding Remarks

Where test suites are appropriate, agents might also help scientists write and automate them, although whether this convention takes hold is an open question.

— How Researchers Use and Verify AI Coding Assistants: Tasks and Validation Practices in Scientific Programming  (2609.22049 - O'Brien et al., 18 Sep 2026) in Section Future research