Turing Award winner Hellman and other experts: AI governance cannot find answers only in code

When machines begin to think, how should humans coexist with them? When algorithms participate in decision-making, how is security guaranteed? At the AI Parallel Forum of the 2026 World Laureates Forum held on October 30, Turing Award laureates, mathematicians, academicians of the Chinese Academy of Sciences, and chip engineers gathered together to confront the same question: as AI is reshaping the world at an unprecedented speed, how exactly should its next step proceed? And where is it heading? The "top minds" gave different answers, but the consensus was very clear: how far AI can go in the future does not depend on how smart the machines are, but on how clear-headed we are. Martin Hellman. Photo by Wen Hui Bao reporter Yuan Jing. AI is replicating risks into more domains. Martin Hellman, winner of the 2015 Turing Award and professor emeritus in the Department of Electrical Engineering at Stanford University, opened with a sharp judgment: "Artificial intelligence, nuclear weapons, and climate change are not the real problems—they are merely symptoms of a deeper problem." In his view, the core contradiction of human civilization has never changed: technology has given us almost "god-like" power, but our behavioral patterns remain at the "adolescent stage." And AI is replicating "runaway risk" into more domains—he cited public media reports: AI can easily teach people to plan violent attacks and modify dangerous pathogens; drones fully guided by AI have already caused real casualties. This scholar, who has worked for half a century in cryptography and global security, is not a doomsayer. He said that he and his wife were once on the brink of divorce, but now they have not quarreled for more than 20 years. "We turned disagreement into an opportunity for collaboration." In his book, he analogizes the logic of repairing intimate relationships to global security: you cannot repair a broken marriage overnight, nor can you step into an ideal world of peace in one move, but we must always work toward that goal. He stressed that change must come from within society—provided that we first see ourselves as "human beings," rather than as separate nationalities, ethnic groups, and positions. "We still have the ability to make changes now, and to survive in the technological age." Roundtable forum. Every large model should come with ethical review. What exactly should humanity do to use AI safely and effectively? Noga Alon, winner of the 2024 Wolf Prize in Mathematics and a professor at Princeton University, pointed out AI's "black box dilemma": "We do not know what is actually happening inside large language models. All we can do now is adjust the weights of the neural network, but no one truly understands why it makes its final decisions." For this reason, he believes AI safety cannot be solved by a single actor; countries and companies must abandon vicious competition and reach collaboration, otherwise "the race itself will breed danger." Chen Yunji, deputy director of the Institute of Computing Technology of the Chinese Academy of Sciences and director of the State Key Laboratory of Processors and Chips, analyzed that the underlying logic of mainstream large language models is statistics, extracting patterns from massive data to predict the next token. "Based on statistical principles, the accuracy rate is always 99%, never 100%. And that 1% uncertainty is enough to cause harm." In his view, the next generation of AI needs to embed physical laws and logical rules into models in order to achieve "zero errors" in key scenarios. Chen Yunji believes that in the next decade, AI is highly likely to achieve "self-improvement": AI models optimizing AI models, and AI chips improving AI chips. Of course, ethical rules must be embedded into technology at the same time. "Every large model should come with an ethical review mechanism; this should become a global consensus." So, will humans be replaced by AI? Yurii Nesterov, winner of the 2023 World Laureates Award and a professor at the Chinese University of Hong Kong (Shenzhen), used a jigsaw puzzle metaphor: AI can efficiently assemble a million-piece puzzle, provided that all the pieces already exist. If one piece is missing, it can assemble nothing; but humans can use imagination to fill in the missing piece. He admitted that it is hard to imagine AI having human imagination. White paper released. The ultimate path of technology must be drawn by humans. "Technology itself is unknowable, and its ultimate path must still be drawn by humans." At the forum, multiple experts mentioned that governance cannot seek answers only in code, but must return to the human dimension. Whether it is the underlying rules of technological R&D, the shaping of capabilities in talent cultivation, or the boundary consensus for industrial implementation, human judgment, human values, and human rights must all be put back at the center. He Jifeng, an academician of the Chinese Academy of Sciences and a lifetime professor at East China Normal University, proposed the concept of trustworthy AI as early as 2017: AI cannot be an unfathomable black box; it must be explainable, verifiable, and traceable, and machine behavior must be aligned with human values. "The quality and limitations of all content produced by AI ultimately depend on the depth of human understanding." Reshaping the education system is also an important part of AI governance. Nesterov said bluntly: "When you grade a student's exam paper answered with AI, what you are actually assessing is their ability to pay for AI, and it has nothing to do with their level of knowledge." Alon said that at Princeton, take-home exams, which were allowed in the past, are now "completely impossible," and starting this year every course has restored in-class quizzes. Liu Shaoying, a distinguished professor at the School of Software Engineering of East China Normal University, insists that when students write papers, they should first write in English themselves, and then ask AI to polish the text and use it for comparative learning; students who insist on relying on AI will be invited to the office to read the paper aloud sentence by sentence and explain the reason for every wording choice.