Artificial intelligence (AI) is a broad umbrella term that refers to a range of technologies that have been deployed for scientific, medical, commercial, and surveillance purposes. AI has various environmental and social concerns that have been described elsewhere. Like computers, AI can be designed, executed, and used in more and less ethical ways.
As of September 2026, this is my current personal policy on AI use, which will continue to evolve as the technology and norms evolve:
Utility: AI is useful, and it is being widely adopted in most professions and in many areas of our lives in ways we cannot ignore because it is useful. In the health sciences, AI has the potential to make science more accurate, efficient, and impactful through machine learning, reproducibility analyses, and automation of coding and other research tasks prone to human error. I believe individual users choosing to not use AI will have little impact on whether AI is adopted societally because it is useful. While I understand individuals’ principles and positive intentions underlying a stance to not use AI, I also view it as an idealistic one that is not based on pragmatic impact – though it can be a worthwhile personal choice for some, especially for those whose professional fields are not widely adopting AI use.
Use of Generative AI Professionally: I do not use AI for drafting research articles or essays. I limit my use of AI tools to cases where AI has a clear benefit (i.e., to reduce error, allow me to solve a problem I wouldn’t otherwise be able to solve, save substantial time, or challenge my views). Some examples of my use include as a tool to get critical feedback, for better understanding and implementing complex statistical methods, structuring and clarifying my own thinking on topics, as a tool for audio transcription, for identifying errors, for code review, and for reproducibility and robustness checks. I currently pay $20 per month for a Claude Pro account. I also occasionally make use of institutionally provided AI services.
Use of AI as an Artist: I do not use AI for image or video generation. I do not use AI in my fiction writing. I occasionally solicit feedback from AI on my art, in most cases in regards to how to distribute and share it or to understand how others might perceive or view it.
Carbon Impact: Estimates attribute 2.2 grams of carbon for the servers needed for training and operations of a single ChatGPT query. At 100 queries per day (a relatively heavy user), this equals around 180 lbs per year. For scale, the average American emits 4,305 lbs of carbon per year from their housing (heat, electricity), waste, and driving, and a single round trip flight ranges from 500 to 5,000 lbs of carbon. I purchase offsets for my annual carbon emissions (including AI use) via Gold Standard, and I have written about this environmental footprint elsewhere.
Water Impact: Similar to how I cannot choose where my electricity comes from that powers my own computer and how clean the source is as an individual consumer, I have little choice or control in the current market over the sustainability of the AI servers that I use. Similar to other issues of consumption, I believe critiquing AI therefore at the institutional, corporate, and societal levels is more fruitful than criticizing how individual users attempt to adapt to a changing technological environment and make use of a complex technology with positive and negative attributes. Many of the environmental and social costs associated with AI could be averted with public policy for clean energy and appropriate regulation of the technology. As more environmentally friendly AI options become available in the future, I will prioritize these, and I politically support increased regulation and monitoring of data centers being constructed for AI expansion.
Other Social Impacts: While I understand the view that AI plagiarizes and steals intellectual property, I also think it has greatly expanded access to knowledge that was previously often gate-kept. Knowledge and ideas are almost always recycled, reused, and adapted from prior knowledge and ideas, and I am personally skeptical of most claims about ‘originality’. In the case of AI, AI has made complex statistical methods accessible to many people which were once only feasible for a few highly privileged scientists to implement. Other social impacts – like labor conditions for people involved in training models – are labor issues that are present in many industries and which warrant policy and other structural / societal action. While AI is resulting in job losses, AI may also present an opportunity for universal basic income. The fact that this is unlikely to happen in the short-term is a policy failure.
In conclusion, personally, I don’t think it will work to just try to make AI go away. I think we need to recognize that it is here to stay, that it needs to be heavily regulated by policymakers, and that it has the potential of delivering societal benefits if we insist on this through policy. I will continue to update this position statement as disciplinary norms evolve and as my own policies develop and change.
