Microsoft CEO Nadella: No Longer Believes That Improved AI Capabilities Will Naturally Solve the Alignment Problem
Microsoft CEO Satya Nadella said on the "20VC" podcast that he previously believed that the more powerful an AI system becomes, the more easily its behavior would conform to human values and goals; he no longer believes the two will develop in natural sync, and said alignment may be a thornier problem of a different nature. He previously argued in an X post on October 10 for strengthening external security controls, believing that companies should treat AI models as internal security threats, design protections on the assumption that models have been compromised, and establish mechanisms through deterministic system design, human controls, and operating procedures.
Why read this
AI score 52
Nadella publicly revises his judgment on the AI alignment problem and offers a protection approach that treats models as internal security threats.
Topics
Body · AI translation
Microsoft CEO Nadella: I used to think the more powerful AI is, the safer it would be — I don't believe that anymore
English text is being prepared.
English text is being prepared.
On the new episode of the 20VC podcast, host Harry Stebbings asked Nadella what beliefs had changed over the past year. Nadella explained that he had originally thought that as AI capabilities continued to improve, the "alignment" problem of ensuring AI behavior conforms to human values and goals would also resolve itself.
The effects brought by scaling up models met Nadella's expectations, and AI capabilities have continued to improve, but the alignment problem remains unsolved. "I always used to think the alignment problem would just solve itself. I don't think that anymore. I think alignment may be a thornier problem, and different in nature."
On October 10, Nadella posted on X warning about the risks of AI and advocating stronger external safety controls. He argued that companies should treat AI models as insider security threats and, when taking protective measures, should assume the models have been compromised.
Nadella wrote in the post: "We need to build safeguards for AI models whose behavior is hard to fully predict, through reliable, deterministic system design, human oversight, and sound operating procedures. For areas where existing standards are insufficient, industry standards also need to be developed."
English text is being prepared.