The intersection of chemistry and artificial intelligence (AI) is a fascinating area that attracts a lot of attention in both research and industry. We talked to people working in the field about the potential of AI to revolutionize chemical research, but also concerns, (current) limitations, and ethical implications for chemical applications. We also asked for ideas to try or experiment with, as well as useful articles and videos for beginners and advanced users.
Thijs Stuyver is a Junior Professor in the Chemical Theory and Modelling group at Chimie Paristech, Paris Sciences et Lettres University (PSL), France, who works in the field of artificial intelligence-supported chemistry.
What fascinates you about AI?
I find the rapid impact of AI on our everyday lives to be truly fascinating. In less than a year, ChatGPT has evolved into my personal assistant, aiding me in a variety of work-related tasks—from coding to suggesting e-mail drafts and even reformatting my references according to the desired journal style for my papers. I leverage AI to generate figures and artwork more efficiently, resulting in visually superior outputs.
With a continuous stream of new and increasingly advanced tools and applications set to be released in the coming months and years, my productivity will likely continue to soar. I am excited about the intellectual bandwidth that will be freed up all across society in the years ahead because of these evolutions, allowing a sharper focus on challenges that truly matter, now that a lot of the menial work is becoming (semi-)automated.
Is there anything we should fear?
As an optimist, I don’t lend much credence to doomsday scenarios in discussions about artificial intelligence. It is a disruptive technology, but humans will adapt, as they have to every other disruptive technology that has been unleashed on the world in the past.
What I do see as a major challenge, particularly within the context of chemistry education, is how we maintain a balance between the new tools and capabilities AI provides us with, and the—equally important—skill set of a traditionally trained human chemist. In an era dominated by flashy headlines and hype, there’s a temptation to prioritize the use and development of machine learning models over gaining a deep understanding. I am fully convinced that machine learning will (continue to) play an indispensable role in chemical research in the foreseeable future, and its impact will only grow. However, there are aspects to scientific discovery that cannot easily be outsourced to a model. Without a profound human knowledge of chemistry, one lacks the reference framework to identify worthwhile topics/problems and to contextualize model predictions and their limitations. Hence, the core of traditional chemistry education will remain highly relevant, and the challenge lies in simultaneously nurturing new digital skills while retaining the essence of what makes a skilled chemist.
Do you have something for our readers to try out or experiment with?
It’s fascinating to witness the rapid advancements in image recognition for chemical reaction schemes. It is somewhat of a niche problem, but in quite a number of my projects, I need to extract reaction data from reaction schemes in (organic) chemistry papers. This used to be a tedious process where I essentially had to re-draw the scheme in ChemDraw and then convert my drawing into a SMILES string.
With new tools such as RxnScribe, I can simply paste an image of a reaction scheme, and the extraction of reactant and product structures, as well as textual descriptions of the reaction conditions, is performed automatically.
Can you recommend a good article/video/website for beginners and one you enjoyed recently?
For a comprehensive introduction to artificial intelligence, particularly deep learning, I highly recommend the pedagogical textbook by Goodfellow et al. The book, available for free download at https://www.deeplearningbook.org, served as my own entry point into the field and continues to be an invaluable resource for insights and understanding.
Additionally, for an accessible introduction to machine learning with a focus on chemistry, I recommend the insightful review by Mater and Coote:
- Deep Learning in Chemistry,
A. C. Mater, M. L. Coote,
J. Chem. Inf. Model. 2019, 59, 2545–2559.
https://doi.org/10.1021/acs.jcim.9b00266
Although a few years old, the article still provides a solid foundation—despite the rapid progress in this dynamic field.
A recent article that I particularly liked is the work by Julia Westermayr and coworkers:
- High-throughput property-driven generative design of functional organic molecules,
J. Westermayr, J. Gilkes, R. Barrett, R. J. Maurer,
Nat. Comput. Sci. 2023, 3, 139-148.
https://doi.org/10.1038/s43588-022-00391-1
To me, this paper stands out as an excellent showcase of the capabilities offered by new data-driven approaches.
Thank you very much for the insights.
back to overview “Opinions on AI & Chemistry”
