If you're interested in coming to @ucddublin.bsky.social School of Chemistry to work on chemical reaction networks, electrochemistry, sustainable chemistry, and/or philosophy of chemistry, please reach out to @ewcspottesmith.bsky.social and express your interest!

Instructions are on our website.

There is also a need to revisit chemical ethics and our codes of conduct to adapt to changing technology (we note that most codes are heavily experiment-focused and ignore #CompChem). There are a lot of open questions, but we hope this opens the door to a much-needed discussion! (10/10)

So, foundation models are ethically fraught, in many cases downright unethical. What are chemists to do? We argue that the first step is to identify where these models are truly necessary, and a) eliminate all unnecessary applications; and b) actively seek alternatives for necessary uses. (9/10)

Finally, we examined chemical foundation models, such as protein models (think AlphaFold/RoseTTAFold) and "universal" ML interatomic potentials. These models suffer from few of the ethical problems of general-purpose models like LLMs, but their environmental cost is still a cause for concern. (8/10)

as @gabegomes.bsky.social recent work has shown, and raises novel ethical risks. The use of foundation models in chemical education is also revealed to be deeply problematic with (in many cases) little clear benefit for students or instructors. (7/10)

We then look at chemistry-specific applications of general-purpose foundation models in four areas: text generation, classification & regression, laboratory automation, and education. Laboratory automation introduces foundation models as dual-use technologies, (6/10)

In terms of chemical professional ethics, the case is less clear. There are some obvious violations (e.g., plagiarism, bias, and environmental harm), but also some ethical requirements that are unaffected or even possibly supported by the judicious use of "AI" tools. (5/10)

With this foundation, we first analyzed generic, off-the-shelf applications of foundation models like LLMs. We find that they violate pretty much all of science's ideals, with objectivity, honesty, and stewardship being the most problematic and accountability being the least. (4/10)

So, we performed a qualitative study of 5 professional codes of ethics and identified chemistry's stakeholders (with the public, future generations, and the environment being most key) and 13 ethical requirements for chemical professionals. This forms a basis for our analysis. (3/10)

First, what ARE the ethics of chemical practitioners? We can look to scientific ideals like those from the US National Academies (objectivity, honesty, openness, accountability, fairness, stewardship), but a) these ideals are only aspirational; and b) they aren't specific to chemistry. (2/10)

A new version of our preprint on the scientific and chemical ethics of "AI" foundation models is now out! This version has a much tighter focus on chemistry and clearer links to specific scientific ideals and chemical ethical norms. #AIethics #CompChem #ChemEthics #GenAI #MLethics