AI Writes 36 Research Papers in 3 Months, Solves 150-Year Puzzle

Scientists just got a new research partner that never sleeps and reads across every field imaginable.

Harvard physicist Matthew Schwartz created BootLoops, an AI tool that hunts for solutions hiding in plain sight across different scientific disciplines. Working with 19 coauthors and experts who verified the findings, the system produced 36 research papers in three months by spotting connections most researchers would never have time to find.

The breakthrough came when Schwartz realized AI could solve what he calls "Claude-shaped problems." He started by getting Anthropic's Claude AI to combine scattered particle physics equations into one framework. The model reproduced results from one of his papers in just 20 minutes.

Then something remarkable happened. Claude itself noticed that equations from physics kept appearing in completely different fields. It found the same mathematical patterns in population genetics, evolutionary biology, linguistics, and even Medicare insurance ratings.

One discovery could shift up to $1.4 billion in Medicare bonus payments between 2025 and 2027. BootLoops found that the shortcut algorithm Medicare uses to rate health plans doesn't actually follow its own rules. Using a statistical method from 1958, the AI calculated the precise boundaries that would push 34 health plan ratings across bonus thresholds.

Another project cracked a 150-year-old mystery about sunspots. Because the sun's rotation hides sunspots from view for most of their lives, scientists never agreed on how long they last or why they disappear. BootLoops borrowed a medical tool doctors use to measure mortality risk and applied it to every sunspot logged since 1874.

The answer surprised everyone. Sunspots age, with their risk of "dying" rising about 20 percent each day regardless of size. None of the existing theories predicted this.

The system isn't perfect. Schwartz admits Claude "loves to declare victory" too early, and many initial findings were technically correct but scientifically boring until human experts steered the research. That's why he brought in specialists from each field to judge which of 400 candidate problems were actually worth pursuing.

The Ripple Effect

The real power here isn't replacing scientists. It's giving them a tireless research assistant that reads across boundaries most humans can't cross. A physics researcher would never spend six months reading linguistics papers hoping to find useful methods, says Georgia Channing from Hugging Face. But AI can make those connections instantly.

BootLoops now spans 22 fields and keeps growing. It's tackled how Earth's atmosphere first filled with oxygen, how quickly tree species change in Panamanian rainforests, and how natural selection removes harmful mutations in humans. Each new problem adds new tools to help solve the next one.

The system proves that thousands of solutions already exist in scientific literature, just waiting in the wrong journals for someone to connect the dots.