Will AI ever win a Nobel Prize?

In 2016, biologist and Sony AI chief Hiroaki Kitano challenged the scientific community to create an artificial intelligence system capable of making a discovery worthy of a Nobel Prize. According to the "Nobel Turing Challenge," by 2050 a system should emerge that can independently formulate hypotheses, design experiments, and analyze data to achieve a scientific breakthrough without human intervention.

Some researchers believe this could happen much sooner. Ross King from the University of Cambridge, one of the challenge organizers, thinks an AI scientist might reach laureate status in just 10 years. Sam Rodriques from the FutureHouse research lab names an even bolder timeline - 2030.

What AI can already do

Over the past two years, AI models have shown impressive results. They help decode animal speech, predict stellar collisions, and optimize quantum computers. The Coscientist system can already independently plan and execute complex chemical reactions using robotic laboratory equipment.

In 2024, the Nobel Prize in Chemistry went to the creators of AlphaFold, a system that predicts protein structures. But this was an award for developing the AI itself, not for discoveries made by it autonomously.

Researchers identify three waves of AI in science. The first is when AI acts as an assistant on specific tasks. The second is when it learns to generate and test its own hypotheses by analyzing data and literature. The third, final stage, is a fully autonomous AI scientist that asks questions, designs experiments, and conducts them without human participation.

Obstacles

Many researchers are skeptical of such predictions. Doug Downey from the Allen Institute for AI found that agents based on large language models successfully complete individual scientific tasks 70% of the time, but manage the full research cycle (from idea to report) only 1% of the time.

The main problem: current AI systems don't understand underlying principles, they merely mimic results. One study showed that a model could predict a planet's orbit but couldn't reproduce the laws of physics governing that process. Another system learned to navigate New York City but couldn't create a map of its streets.

Subbarao Kambhampati from Arizona State University points to a fundamental limitation: AI experiences the world only through data, unlike humans with their lived experience.

Dangers of Automation

Some scientists question whether this goal is worth pursuing at all. Anthropologist Lisa Messeri at Yale University and psychologist at Princeton University Molly Crockett warn that over-reliance on AI is already introducing more errors into science. They fear that scientists will begin to "produce more but understand less."

There's also a social concern: AI performs tasks that once trained young researchers. If machines replace this learning stage, future generations of scientists may never acquire the skills necessary for their own discoveries. "Given the shrinking of research budgets, we are at a concerning moment for evaluating the pros and cons of this future," says Messeri.

Earlier, Kazinform News Agency reported that AI could boost global trade by nearly 40% by 2040.