Updated: 24 Feb 2026
This old dog is trying some new tricks — or more accurately, I am newly trying some old tricks, that characterized my early academic career which focused on math and coding.
I am now experimenting with using AI in support of my research.
Early in my career, when I was a staff scientist at the U.S. National Center for Atmospheric Research, I did a lot of quantitative work that might be classified as old-school climatology (see, for example this paper on precipitation, floods and damage, and this one on hurricanes and climate change).
AI tools — particularly for quantitative analysis and coding — provide an opportunity to have a set of AI assistants in support of my work, allowing me to dive deeper and more efficiently into more technically complex topics that are time intensive when performed by us mere humans.
My approach is to use AI much as I would student assistants, grad students, and post-docs back when I was a professor. The goal is higher quality work produced more efficiently.
Specifically, I currently use ChatGPT for basic research support (e.g., “find me the flowchart mentioned in this government document”) and Claude for quantitative applications (e.g., please digitize this figure, calculate the Mann-Kendall trend for variable X, and produce data and code for replication). As longtime THB readers know, I’ve long used AI to generate cartoons and header images for my posts — that continues. I pay for each service.
So far I am impressed — more than that, amazed.
AEI, where I am a Senior Fellow, has an AI policy for its scholars which states:
[S]cholars should not use LLMs to draft their publications or to generate the arguments, ideas, and evidence that scholars go on to use in their publications. This is the same standard the Institute has long applied to scholar use of research assistants or reference works. At the same time, scholars should feel free to use LLMs as part of the research and writing process, in the same way that they use RAs and references: To understand the state of the literature or debate to which they are contributing, for copy editing, to find important scholarly work that is related to their project, and for general education.
This is a solid policy, and I support it 100%. For me, writing is thinking, and that can’t be outsourced.
I haven’t yet used AI for copy editing, but that changes today. What have I been waiting for? For the THB typo finders, who I have much appreciated, you’ll now be out of business!
If and when my approach changes I’ll update this page. In the meantime, expect some productivity gains here at THB. I’d welcome hearing about your experiences and your recommendations for the effective use of AI in research.
Thanks for reading!