AI for Science Really Means Engineering for Science

October 8, 2024

Steve Crossan’s article Engineering for Science asks:

“Why did AlphaFold happen at DeepMind rather than (for example) the Broad Institute ? It wasn’t data. Everyone had access to exactly the same data. It wasn’t compute. The compute budget for AlphaFold1 was well within the budget of an academic project. The real reason was that we treated it as an engineering problem as much as a research one. In fact, this was the secret sauce of DeepMind: around ⅓ of the overall headcount was devoted to what we called Research Engineering.”

Indeed! 😊 Glad to have been one of those Research Engineers working in the beginning of AlphaFold, alongside Marek Barwinski, Richard Evans, Laurent Sifre, James Kirkpatrick and Andrew Senior. Steve Crossan got the AlphaFold project officially started, after the initial Hackathon results, in early 2016. We had help from other engineers, such as Vedavyas Panneershelvam, for things like using the Rosetta library at Google, compile it with blaze, etc. We had to set up part of the infrastructure in Google cloud, write python wrappers, write a leaderboard with automated metrics, etc. Lots of “boring” engineering work. The foundation for later successes.

Now at Inductiva.AI we noticed the same pattern: we envision a world where physics simulations will merge with Machine Learning, but there is a lot of infrastructure work that needs to be done first, and that’s mostly software engineering. Our API will lower the barrier of entry for scientists and engineers to generate large Physics datasets to train ML models in different fields (fluids, molecules, materials, etc). Doing the “boring” work first. ;)

Note: this was a reaction to Steve Crossan’s article, written the day before the announcement of the Nobel Prize for AlphaFold - little did we know!