Recap: CAFE’s State of the Science Webinar on AI and Data Science
Thank you to all who joined CAFE’s State of the Science: AI and Data Science - Beyond the Buzzwords webinar with Dr. Marianthi Anna Kioumourtzoglou from Brown University and Dr. Joan Casey from the University of Washington! If you missed it, or want to review key points, here’s a recap, with links to the relevant moments in the recording.
How Researchers Are Currently Using AI
Dr. Casey outlined practical ways environmental health researchers are engaging with generative AI tools:
- Coding assistance
- Assisting in literature review searches
- Editing text for manuscripts/proposals
- Generating graphs for presentations
The shift to using AI has been dramatic. Stack Exchange, once the go-to resource for coding help, has seen a 75% drop in traffic since 2017 as researchers have begun to turn to AI tools instead.
Concerns
Electricity Consumption: Data centers now consume approximately 4% of U.S. electricity, projected to reach 6.5% - 12.5% by 2028. About 56% of this electricity comes from fossil fuels, and communities near data centers are paying 100 - 200% more for electricity compared to 5 years ago.
Water Consumption: Data centers are water-reliant. Microsoft saw a 34% increase in water consumption from 2022 to 2023, with AI estimated to use more water annually than the country of Denmark by 2027.
Algorithmic Bias: AI that is trained on biased or incomplete data will reproduce and amplify those same inequalities in its results. If some communities are underrepresented in the training data, AI models may produce findings that favor well-represented groups while misrepresenting marginalized populations.
Overconfidence in Wrong Results: AI can generate code that runs and produces results but is methodologically wrong. As Dr. Kioumourtzoglou noted, students may receive statistically wrong answers without recognizing the error, which is a risk for those without strong methodological training.
Key Takeaway
Dr. Kioumourtzoglou and Dr. Casey noted that AI can complement but not replace rigorous epidemiological training. While these tools offer tremendous potential for handling complex environmental health data, researchers must maintain critical thinking and push for regulations addressing AI’s environmental impacts.
