What science looks like when the silos come down

For most of my career, simulation lived in one building, data analysis lived in another, and AI lived somewhere down the road, mostly in computer science departments. People moved between them, but the methods didn’t, not really. You did one of the three and borrowed the others when you had to.

That arrangement is coming apart. The interesting science problems now want all three at once, in the same workflow, on the same machine, often inside the same job. Not because anyone declared it should be that way, but because the problems don’t sit cleanly inside any one of the categories anymore. The boundary between training a model, running a simulation, and analyzing the result is getting fuzzy enough that drawing it isn’t useful.

The GenSLMs project is a good example, which is part of why it’s worth talking about. The team trained genome-scale language models on raw nucleotide sequences to track how SARS-CoV-2 was evolving. Biology, language models, and leadership-class computing in one pipeline. My role was on the ALCF side, helping to enable the runs. The 2022 ACM Gordon Bell Special Prize for COVID-19 Research, announced at SC22 in Dallas, recognized the work, and the team behind it across Argonne, the University of Chicago, NVIDIA, Cerebras, UIC, Caltech, Harvard, NIU, the Technical University of Munich, and the ALCF.

The reason the prize matters more than the usual press release is that it’s the community putting the new shape on the record. A Gordon Bell isn’t given for projects that look like everything else; it’s given for projects that look like where things are going. That this one combined simulation, data, and AI on a single problem, and won, says something about which direction the wind is blowing. The next decade is going to look more like this and less like the one before it.

Coverage of the award is at Inside HPC.