Scientists predict the possibility of artificial intelligence winning the Nobel Prize

In recent years, artificial intelligence (AI) has demonstrated its ability to analyze data, design experiments, and come up with new scientific hypotheses, leading many researchers to believe that AI could one day rival the most brilliant minds in science—even making Nobel Prize-worthy discoveries, according to Nature magazine.

In 2016, biologist Hiroaki Kitano, CEO of Sony AI, initiated the “Nobel Turing Challenge” – a call to develop an AI system smart enough to make a Nobel-level scientific discovery on its own.

According to the project's goal, by 2050, an “AI scientist” will be able to form hypotheses, plan experiments, and analyze data without human intervention.

Researcher Ross King, University of Cambridge (UK), believes that milestone may come sooner: “It is almost certain that AI systems will reach the level of winning Nobel Prizes. The question is only in the next 50 years or 10 years."

However, many experts are cautious. According to them, current AI models rely mainly on available data and knowledge, and do not really create new understanding.

Researcher Yolanda Gil (University of Southern California, USA) commented: "If tomorrow the government invests one billion USD in basic research, progress can accelerate, but it is still very far from that goal."

To date, only people and organizations have been awarded Nobel Prizes. However, AI has contributed indirectly: in 2024, the Nobel Prize in Physics went to pioneers in machine learning; that same year, half of the Chemistry Prize went to the team behind AlphaFold, Google DeepMind’s AI system that predicts the 3D structure of proteins. But these prizes honor the AI’s creators, not the AI’s discoveries.

To be worthy of a Nobel Prize, according to the Nobel Committee's criteria, a discovery must be useful, have a far-reaching impact and open up new directions of understanding. An “AI scientist” who wants to meet this requirement must operate almost completely autonomously - from asking questions, choosing experiments to analyzing results.

In fact, AI is already involved in almost every stage of research. New tools are helping to decipher animal sounds, predict collisions between stars, and identify immune cells vulnerable to COVID-19.

At Carnegie Mellon University, chemist Gabe Gomes' team developed “Coscientist” – a system that uses large language models (LLMs) to autonomously plan and execute chemical reactions using robotic devices.

Some companies like Sakana AI in Tokyo are looking to automate machine learning research using LLM, while Google is experimenting with chatbots that collaborate in groups to generate scientific ideas.

In the US, the FutureHouse lab in San Francisco is developing a step-by-step “thinking” model to help AI ask questions, test hypotheses and design experiments – aiming for a third generation of “AI in science.”

The final generation will be AI that can ask questions and conduct experiments on its own, without human supervision, according to FutureHouse director Sam Rodriques. “AI could make Nobel Prize-worthy discoveries by 2030,” he predicts. The areas with the most potential are materials science and the study of Parkinson’s or Alzheimer’s disease.

Other scientists are skeptical. Doug Downey of the Allen Institute for AI in Seattle says a test of 57 “AI agents” found that only 1% could complete a research project completely—from idea to report. “Automated scientific discovery from start to finish remains a huge challenge,” he says.

In addition, AI models still do not fully understand the laws of nature. One study found that a model could predict planetary orbits but not the underlying physical laws; or could navigate a city but not create an accurate map.

According to expert Subbarao Kambhampati (Arizona State University), that shows that AI lacks the real-life experience that humans have.

Yolanda Gil argues that to reach Nobel status, AI needs to be able to “think about thinking” – that is, to self-evaluate and adjust its own reasoning processes. Without investing in this foundational research, “Nobel-worthy discoveries will remain a long way off,” Gil says.

Meanwhile, some scholars warn of the dangers of over-reliance on AI in science. A 2024 paper by Lisa Messeri (Yale University) and Molly Crockett (Princeton University) argues that overuse of AI could increase errors and reduce creativity, as scientists “produce more but understand less.”

“AI could deprive young scientists who might otherwise win big prizes in the future of learning opportunities,” Messeri added. “With research budgets shrinking, it’s a worrying time to consider the cost of that future.”

(TTXVN/Vietnam+)

Source: https://www.vietnamplus.vn/gioi-khoa-hoc-du-doan-kha-nang-tri-tue-nhan-tao-gianh-giai-nobel-post1068525.vnp