Will AI ever win its own Nobel? Some predict prize-worthy science discovery soon (www.nature.com)

AI systems are increasingly moving from tools to active scientific agents, and some researchers now argue they could make discoveries worthy of a Nobel prize within decades. The Nobel Turing Challenge, proposed in 2016, frames the goal: an AI scientist that autonomously generates hypotheses, designs and runs experiments, and interprets data to produce breakthroughs “fully or highly autonomously.” While AI has already shaped award-winning science indirectly (e.g., AlphaFold and neural-network pioneers honored in 2024), no machine has yet met the autonomy threshold required for its own prize. Optimistic timelines range from around 2030 to 2050, with proponents like Ross King and Sam Rodriques predicting rapid progress if research and funding intensify. Technically, progress rests on advances in large language models (LLMs), “reasoning” models trained on stepwise problem solving, and lab automation. Demonstrations such as Coscientist — an LLM-driven system that plans and executes chemistry workflows with robotics and fast computational chemistry — and LLM agents that mine papers and datasets to surface overlooked biological insights show capability growth. Key hurdles remain: LLM hallucinations, the need for human oversight in many workflows, reproducibility and safety risks, and the sociotechnical question of attribution (current prizes go to living humans/institutions). If models can reliably self-generate valid hypotheses, design experiments, and navigate wet-lab risks, fields with large open problems (materials science, neurodegenerative disease) are prime candidates for an AI-driven, prize-worthy discovery.