Summary of Key Points
Google DeepMind has disbanded the AlphaFold team, which achieved Nobel Prize-level breakthroughs, and integrated it into the Gemini-driven “AI for Science” initiative. This marks not a rejection of specialized models but an evolution in the approach to AI research: shifting from using dedicated models to solve specific scientific problems to involving general-purpose intelligent agents in the entire research process. The three leading North American AI companies (Google, OpenAI, and Anthropic) are all investing in this direction, which has sparked a competition for top scientists. China and the United States are also engaging in national-level competition in this field.
1. The Disbandment of AlphaFold is Not a Failure; It Represents a Change in AI Research Methods
AlphaFold represented the pinnacle of AI’s ability to solve single scientific problems (such as predicting protein structures), but it was limited to that task. Now, DeepMind aims to transform AI from a tool into a research assistant that can not only solve problems but also read literature, formulate hypotheses, design experiments, and analyze results on its own. The former members of the AlphaFold team have been assigned to new projects such as the Gemini intelligent agent and enzyme design, with the goal of turning specialized models into “skill modules” that can be invoked by the AI agent—similar to how a calculator app in a smartphone is a function that can be used by an intelligent assistant like Siri, rather than existing independently.
2. North American Giants are Developing “All-Around AI Research Assistants”
Google has released the Gemini for Science suite, which includes AI Co-Scientist, capable of generating hypotheses, and AlphaEvolve, which optimizes algorithms. OpenAI is integrating GPT into laboratories to conduct experiments and may have even solved a mathematical conjecture that has stumped scientists for 80 years. Anthropic’s Claude Science can create multiple “sub-AIs” to perform different tasks (e.g., one for literature research and another for data analysis). Their common goal is to make AI capable of covering the entire research process, not just a single step.
3. The Competition for Talent is Changing: AI Companies are Now Competing for Scientists
In the past, AI companies competed for machine learning engineers; now, they are competing for scientists with knowledge in physics, chemistry, and biology. For example, the key figure behind AlphaFold, Jiang Pei, has joined Anthropic, and a Fields Medalist (the highest award in mathematics) immediately joined OpenAI after receiving her award. Why? Because AI research requires scientists to provide “real-world feedback”—for instance, to verify whether the hypotheses proposed by AI are correct and whether the experimental designs are sound. Scientists can help AI improve its performance through this feedback loop. ByteDance has also launched a STEM scientist program with a similar approach.
4. AI in Research Must Meet Two Critical Requirements: Reliability and Security
AI in research cannot just generate random answers like chatbots; errors must be identified during experiments. Therefore, three requirements must be met: ① The AI must not provide misleading information (no “delusions”); ② The process must be auditable (with clear records of data sources and experimental steps for human review); ③ The AI must be able to generate new questions, not just respond to known ones. There are also security concerns: if AI can design drugs, it could also create harmful biochemical weapons, so a “controlled environment” (accessible only to authorized personnel) and safety verification systems are necessary.
5. China and the United States in a New Competition Arena: AI for Science as a National Level Contest
The White House has elevated the use of AI in science to a national competitive priority, aiming to leverage existing models and research strengths to transform technological leadership into long-term scientific discovery capabilities. Chinese companies are also making progress (e.g., ByteDance’s initiatives). The key to future competition will be who can establish a complete ecosystem consisting of models, intelligent agents, experimental platforms, and scientists—this is not just a race for technology but also a contest of research systems.
In One Sentence
The way AI conducts research is shifting from achieving breakthroughs in individual areas to participating in the entire process. Giants are competing for technology and talent, with China and the United States vying for dominance in future scientific research. This is not the end of AlphaFold but the beginning of a new era where AI will transform science.