10 Years, One Mystery—and a 48-Hour AI Revelation

For the last ten years, a dedicated team at Imperial College London had thrown themselves into one of the most urgent battles in modern medicine: figuring out the mechanism that allows certain bacteria to become resistant to antibiotics. The topic of ‘superbugs’—those stubborn bacteria that can shrug off our best medicines—has long been a scientific headache, and these researchers had devoted a decade to inching closer to an answer.

But the story took a twist worthy of a science fiction novel (with a modern tech giant cameo). The researchers put Google’s scientific AI assistant, Co-scientist—a multi-agent tool built on Gemini 2.0—to the test. Their question for the AI? The very same one that had cost them a whopping ten years of inquiry. The result: Co-scientist landed on the identical hypothesis in just 48 hours. Yes, you read that right: not ten years, but two days.

Surprise, Suspicion, and a Dash of Paranoia

José R. Penadés, a member of the London-based team, didn’t quite believe what he was seeing. To the point that he sent an email to Google asking if the company somehow had access to his computer, since he hadn’t even published his findings yet! Thankfully for digital privacy—and perhaps Penadés’ peace of mind—the answer was a resounding no. The AI had figured it out all on its own, using the information available to it.

Does This Mean AI Can Replace Years of Scientific Labor?

Let’s not get ahead of ourselves: a chunk of that ten-year period was spent proving the validity of the hypotheses, a step that—even in its digital glory—AI can’t exactly speed through. Experimentation in the real world still takes time. But, as the researchers have candidly admitted, if they’d had access to that critical hypothesis from the get-go, they might have saved years of work. This is where AI’s superpower comes into play: not just confirming old ideas, but sparking new ones.

The AI didn’t stop at their main question either. The tool offered up four additional, plausible hypotheses—one of which the team hadn’t even considered, and which they now plan to investigate.

AI and the Future of Science: Threat or Opportunity?

Sure, AI tools sometimes have a knack for making things up (the dreaded ‘hallucinations’). But when it comes to sifting through mountains of data—scientific papers, datasets, archives—AIs like Co-scientist have an undeniable edge. Where human researchers can take years connecting the dots, an AI can spot patterns in hours, giving scientists a powerful shortcut to focus on what really matters.

Inevitably, some people fear that artificial intelligence could threaten jobs or take the “human” out of scientific discovery. But others, including José R. Penadés, see it as a tremendous boost to research, not a replacement:

“I think this is going to change science, that’s for sure,” he said. “I’m witnessing something spectacular, and I’m very happy to be part of it.”

Perhaps the robots won’t be stealing lab coats just yet—but they might be giving our best scientists a much needed turbo boost for the discoveries of tomorrow.