Why visualization still matters in the age of AI

It is fashionable, at the moment, to assume that AI will eat every step of the scientific pipeline. Generate the candidates, score the candidates, pick the winner, write the paper. I don’t think the middle of that sentence holds up under examination. At some point a person still has to look at the result, decide what it means, share it with a colleague, and figure out the next move. That step is not going away, and it is not the part the model is good at.

Which is why visualization, the unglamorous part, still matters. When an AI-driven drug discovery pipeline produces hundreds of thousands of candidate molecules, you cannot read them. You cannot make a spreadsheet of them. The number of candidates is the entire problem. What you need is a way to see the structure of the space, find the regions worth attention, and hand the interesting bits to a chemist who knows what to do with them. Without that step, the candidates may as well not exist.

ChemoGraph, the paper our team had recognized at EuroVis 2023 with an Honorable Mention in the Best Paper category, is a clean example of that point. The work was led by Bharat Kale, then finishing his PhD, and was the core of his dissertation. The team included Austin Clyde, Maoyuan Sun, Arvind Ramanathan, Rick Stevens, and me. It was largely an Argonne effort, anchored in the lab’s COVID-19 drug screening campaigns, and built to make those screenings legible to the people doing the screening.

The award is great, and Bharat earned it. The reason the project is worth talking about beyond the award is that it sits at the seam where AI hands off to humans. As long as that seam exists, and I don’t think it’s going anywhere soon, visualization is doing real work in the pipeline, not decorating it.

The paper is published in Computer Graphics Forum and is available at Wiley Online Library.