Turing Award Winner: AI Autonomy Must Match Verified System Reliability
On October 27, 2026, Joseph Sifakis, the 2007 Turing Award winner and founder of Verimag laboratory, gave an interview to Sina Finance on the sidelines of the Sustainable Global Leaders Conference, discussing AI reliability, risks of AI agents, cognitive offloading, and collaborative artificial intelligence.
Speaking about China's AI development, Sifakis said China is one of the few countries capable of competing with the United States in artificial intelligence. He suggested China develop its own AI vision with a focus on industrial AI, citing its strong industrial base and accumulated industry-specific data as advantages. He also recommended that China work with the European Union on AI regulation and standards.
On AI agents, Sifakis warned that errors can propagate across multiple execution steps in multi-agent systems. He gave an example: if a system relies on 10 sequential steps, each with a 99 percent success rate, the overall success rate drops to about 90.4 percent. He stressed that the autonomy of AI must match its verified reliability. For low-risk scenarios, more autonomy may be allowed, but tasks involving physical environments and potential serious consequences require strict guarantees. He said AI governance must answer three questions: Is the system reliable enough? Can humans control it effectively? Who is responsible when problems occur?
Sifakis also raised concerns about "cognitive outsourcing," saying young people who use AI to complete all their homework may skip the thinking and analysis that are essential to education. He advocated collaborative AI, in which AI acts like a tutor that helps students think rather than simply providing answers, and avoids excessive flattery. He noted that understanding a student's thinking process and adjusting help accordingly is far more difficult than generating answers.
