In the fall of 2024, the Nobel Prize Committee made a decision that sent shockwaves through the scientific community. They awarded half of the Chemistry Prize to Demis Hassabis and John Jumper for developing AlphaFold, an AI model that solved the 50-year-old protein structure prediction challenge in biology. AlphaFold can predict the complex structures of approximately 200 million known proteins, and over two million people worldwide have used this tool.

That same year, Geoffrey Hinton and John Hopfield from the machine learning field received the Physics Prize. In the same year, three scientists closely tied to Google won two Nobel Prizes, marking the first time “AI”—a keyword once confined to computer science conferences—made such a high-profile entrance into the halls of humanity’s highest scientific honors.

But a thought-provoking question arose: Was AlphaFold’s prize a victory for AI itself, or for human scientists leveraging AI as a tool? In other words, how close is AI to independently contributing a Nobel Prize-level achievement?

01. From “Capable Tool” to “Autonomous Discoverer”

To answer this, we must first understand the evolving role of AI in research. AlphaFold is essentially a super predictor—human scientists proposed the challenge of “predicting protein structures,” and AI efficiently handled the computations.

It undoubtedly transformed research efficiency, turning years of work into minutes, but the critical scientific questions, research directions, and final judgments remained firmly in human hands.

The real shift came in 2026. That year, Nature published two landmark studies simultaneously, marking AI’s evolution from a “predictor” to a “discoverer.”

Google DeepMind’s Co-Scientist and FutureHouse’s multi-agent system Robin both adopted architectures where multiple autonomous AI agents collaborated, enabling them to formulate hypotheses, design experiments, and analyze data. Dubbed “AI co-pilots for scientists” by the media, Robin fully automated the process from hypothesis generation to data analysis, independently discovering and validating a candidate treatment for dry age-related macular degeneration. It proposed enhancing retinal pigment epithelial cell phagocytosis as a therapeutic strategy and identified two drugs showing efficacy in vitro—all hypotheses, research directions, data analysis, and visualizations were autonomously generated by Robin.

Here’s a fact worth pausing over: A paper published in Nature featured core scientific contributions from an AI system.

02. AI Is Already “Discovering”—And Not Just in Biology

If Robin’s story seems confined to biomedicine, looking at mathematics and physics reveals an even more staggering picture. In mathematics, DeepMind’s latest AI agent, AlphaProof Nexus, achieved large-scale breakthroughs on research-grade problems, solving nine long-standing Erdős open problems, proving 44 mathematical conjectures from OEIS (the Online Encyclopedia of Integer Sequences), and even cracking the 15-year-old pure O-sequence log-concavity problem.

In China, the AI4Math team led by Peking University’s Professor Bin Dong used a self-built automated AI framework to solve an open problem in commutative algebra—the Anderson conjecture—and completed roughly 19,000 lines of formal verification in Lean 4. This marked the first time an AI framework in China solved an open problem in commutative algebra and achieved large-scale formal verification.

In physics, by late 2025, an international research team used deep learning to identify a new double-Λ hypernucleus after 25 years—the first such exotic atomic nucleus observed with AI assistance.

That same year, an MIT-led team developed an AI technique to generate over 10 million candidate materials with Archimedean lattice characteristics, successfully synthesizing two previously undiscovered compounds whose actual properties closely matched AI predictions.

Any one of these achievements would warrant publication in a top-tier journal. Together, they point to a deeper trend: AI is no longer just “calculating faster” for humans. It’s posing questions humans hadn’t considered, finding connections humans missed, and discovering phenomena humans didn’t anticipate. In 2025, Nobel Chemistry laureate Omar Yaghi delivered a striking speech three weeks after his award, heralding the arrival of an “era of self-evolving science.”

Yaghi’s team used ChatGPT to analyze thousands of synthesis reports, having seven AI agents collaborate on experimental design, literature review, and data analysis. They autonomously designed and optimized the crystallization process for COF-323, a material long deemed extremely difficult to synthesize. His forward-looking summary: “We no longer need to ask what AI can do for science. We must now ask: What can science become when driven by AI?”

03. Why the Nobel Committee Still Hesitates

If AI has already produced original scientific discoveries across multiple fields, why hasn’t the Nobel Committee awarded a prize to an “AI-led” achievement? The first obstacle lies in the Nobel’s own evaluation logic.

The Nobel Prizes reward discoveries or inventions that “confer the greatest benefit to humankind,” requiring the committee to confirm the core contributions of awardees. When a discovery is made jointly by human scientists and AI systems, the question of “who made the critical intellectual contribution” becomes exceptionally thorny.

When the Royal Swedish Academy of Sciences awarded the 2024 Chemistry Prize to Hassabis and Jumper, the citation emphasized their “development of the AI model,” not that “the AI model made the discovery”—a subtle wording choice reflecting the committee’s caution toward AI.

A deeper barrier lies in defining autonomy. In September 2025, Anthropic announced that its large language model Claude, with about 950 agents running simultaneously, discovered a CRISPR-like enzyme system in just 21.5 hours.

This finding sparked debate in the scientific community. Stanford professor Le Cong offered an analogy: AI is like sifting through sand on a beach to find something shiny. Whether it’s glass or a diamond can only be determined in the lab.

In other words, AI can efficiently generate candidate hypotheses and discoveries, but the journey from “interesting candidate” to “confirmed scientific truth” still requires experimental validation and human judgment.

However, these barriers are eroding rapidly. A May 2026 Nature editorial noted that fully autonomous AI researchers have not yet emerged, but multi-agent systems have taken a “critical step” in drug discovery.

The research published by Nobel laureate Jennifer Doudna’s team in Science in July 2026 was even more symbolic. They used AI-designed gene-editing protein SynTnpB, with some variants matching or exceeding the activity of natural enzymes, achieving editing rates up to 1.8 times higher than wild-type. This marked the first experimental confirmation of an AI-designed RNA-guided nuclease structure. Doudna, a 2020 Nobel Chemistry laureate, choosing AI for core design work sent its own signal.

04. Timeline: The Race Between Predictions and Reality

On when AI might independently contribute a Nobel-worthy achievement, the industry offers increasingly aggressive timelines. Anthropic co-founder Jack Clark predicted in a May 2026 Oxford University lecture that AI would assist humans in making Nobel-level scientific discoveries within 12 months.

OpenAI Chief Research Officer Mark Chen set the window at two years, specifically highlighting biology as the most promising field.

Anthropic CEO Dario Amodei called 2027 the “year of the tipping point,” believing “genius-level” Nobel-worthy AI systems would emerge in data centers. These predictions may be overly optimistic, but the underlying technological trends are real.

Nobel laureates adopt a more cautious yet hopeful stance. 2021 Nobel Chemistry laureate David MacMillan stated at the September 2026 Nobel Dialogue in Seoul that AI would enable research previously deemed impossible, but “the qualities required of human scientists will remain unchanged.”

2006 Nobel Physiology or Medicine laureate Craig Mello was more direct: “Many mysteries remain undiscovered in the world, and AI will bring them to the forefront.”

Notably, the 2026 Nobel Prizes in natural sciences were still awarded exclusively to “purely human” achievements, but the committee’s stance is softening.

Awarding the 2024 Physics Prize to Hinton and Hopfield essentially rewarded “foundational work that made AI possible”; the Chemistry Prize to AlphaFold rewarded “revolutionary AI applications in science.”

Following this logic, when the next breakthrough—autonomously discovered by an AI system and rigorously experimentally validated—emerges, the committee will face an unprecedented choice: to award or not to award?

05. The Real Question Isn’t “Can It?” but “How Do We Count It?”

Returning to the original question: How far is AI from independently contributing a Nobel Prize-level achievement? Technologically, the answer is clearer: autonomous discovery capabilities are taking shape, validation loops are accelerating, and the timeline may be two to three years.

But institutionally and philosophically, the answers are far more complex. Historically, Nobel Prizes in natural sciences have always been awarded to “humans.”

When the core intellectual contribution of a discovery comes from a non-human system, the seemingly simple question of “who deserves the reward” will force the scientific community to rethink definitions of creativity, boundaries of contribution, and humanity’s unique role in knowledge production.

Yaghi said something in that speech that perhaps best encapsulates this era: “Science has always been a dialogue between nature and the human mind. Now, with artificial intelligence, we are giving science the ability to think, reason, and evolve on its own.” This statement carries both ambition and unease.

How far is AI from a Nobel Prize? The answer may lie not in AI itself, but in whether humanity is ready to embrace a new era of science where AI can also “make great discoveries.”

— The End —