When AI matched 10 years of research in 48 hours
Originally posted on LinkedIn, March 21, 2025.
A team led by Prof. José Penadés at Imperial College London spent 10 years trying to understand how Superbugs become resistant to antibiotics. Then they gave a short description of the problem to a new Google AI tool from DeepMind, built to accelerate scientific discovery with AI. Within 48 hours, it reached the same conclusion they had spent a decade working toward.
The AI did more than reproduce their discovery. It proposed 4 additional hypotheses, including one the researchers had never considered. Prof. Penadés was so shocked that his first reaction was, "How can this be?" He thought Google had somehow accessed his computer. It had not. This was the computational capacity of AI.
"This is going to change science, without a doubt," he said. Working with the tool felt like "finally playing in the Champions League." The AI's suggestions led the team to open new research directions.
These tools are not broadly available yet. Tools such as AlphaFold are available to some researchers and focus mainly on decoding protein structures, which is also highly relevant to understanding bacteria. Scientists can use open versions of AlphaFold, but most of DeepMind's advanced tools are still unavailable to the public.
This points to scientists becoming more like strategists while AI handles more of the experiments and conclusions. I see the same shift in software development as AI-based development tools become more popular.
