A research team from the University of Cambridge, led by Professor Ross King, has demonstrated a groundbreaking method using artificial intelligence (AI) to discover new cancer treatments. They harnessed the capabilities of GPT-4, a large language model (LLM), to identify unconventional drug combinations that might prove effective against breast cancer. The study, published in the Journal of the Royal Society Interface in 2025, involved a novel closed-loop system where AI-generated hypotheses were iteratively refined and validated by human scientists. The Experiment and its Methodology The researchers prompted GPT-4 to analyze vast amounts of scientific literature to find hidden patterns and suggest new drug combinations. Their criteria were stringent: the AI had to propose inexpensive, safe, and regulator-approved drugs that could target cancer cells without harming healthy ones, and exclude standard cancer treatments. This approach aimed to explore drugs typically used for other conditions, such as high cholesterol and alcohol dependence, for their potential in cancer therapy. Initial Findings In the first phase of the experiment, GPT-4 proposed 12 drug combinations. These suggestions were rigorously tested in a laboratory setting against a common breast cancer cell line. Surprisingly, three of these combinations performed better than existing breast cancer drugs, indicating a strong therapeutic potential. Encouraged by this success, the researchers fed the results back to GPT-4, which then generated four additional combinations. Three of these new pairs also showed promising results in subsequent tests. Drug Combinations Identified Among the identified drug pairs, simvastatin (a cholesterol-lowering medication) and disulfiram (an alcohol dependence treatment) emerged as particularly effective against breast cancer cells. These findings suggest that certain drugs, when combined, can exhibit enhanced cancer-fighting properties despite not being traditionally used in oncology. The researchers highlight that these drugs could be repurposed for cancer treatment, though they would require further clinical trials to confirm safety and efficacy. Iterative Collaboration The study demonstrated the power of a collaborative scientific approach, where AI and human researchers work together in a continuous feedback loop. Human scientists provided expert guidance and interpreted the AI's outputs, while GPT-4 rapidly generated and refined hypotheses. This method allowed for the exploration of subtle synergies and overlooked pathways in drug interactions, accelerating the discovery process. Addressing AI Hallucinations GPT-4, like other LLMs, is prone to generating false or inaccurate results, often termed "hallucinations." However, in the context of scientific research, these hallucinations were leveraged as a source of creative and unconventional ideas. The researchers carefully examined the mechanistic reasons behind each AI-suggested combination, ensuring that any misleading information could be cross-verified and corrected through experimentation. Industry Insights and Implications Industry experts have praised the study for pushing the boundaries of AI in healthcare and drug discovery. Dr. Hector Zenil from King's College London emphasized that supervised LLMs offer a scalable and imaginative layer to scientific exploration, enhancing human capabilities rather than replacing them. He noted, "The AI functioned like a tireless research partner, rapidly navigating an immense hypothesis space and proposing ideas that would take humans alone far longer to reach." Professor Ross King echoed this sentiment, highlighting the ability of LLMs to bridge disciplinary gaps and incorporate prior results, marking a significant advancement in the field of scientific research. Company Profiles and Support The research was partially funded by the Alice Wallenberg Foundation and the UK Engineering and Physical Sciences Research Council (EPSRC). These organizations support cutting-edge research in various scientific domains, underscoring the importance and potential of this AI-driven methodology in drug discovery. The study's success opens new avenues for further research and development, potentially leading to more affordable and innovative cancer treatments. This collaboration between AI and human researchers represents a paradigm shift in the scientific method, demonstrating the potential for AI to enhance and expedite the discovery process in complex fields like oncology. As AI continues to evolve, its integration into scientific research holds promise for uncovering new therapies and advancing medical knowledge.