Asimov's three laws don't work: cryptography pioneer Whitfield Diffie on killer drones, AI guardrails and human control

The 1942 robotics laws are useless against autonomous drones and large language models. Diffie: the first law is violated, the second is bypassed by guardrails, and the third is just odd. We use machines as tools for one group of people to control another.

As reported by CCTV+, the three laws of robotics proposed by science fiction writer Isaac Asimov in 1942 are insufficient for addressing today's challenges in artificial intelligence, said Whitfield Diffie, a Turing Award laureate and pioneer of public-key cryptography. In an interview with China Global Television Network (CGTN), Diffie noted that current discussions on AI safety focus too much on regulating what systems produce rather than protecting human rights, stressing that people should not allow AI to make decisions on their behalf.

"First, how little it describes — these ideas are 90 years old — and how little they describe what we do today. But the first law says that a robot may not injure a human being. That's principle number one, and as you'll notice in the case of drones that fly out and kill people, I mean, people argue about whether drones should kill people autonomously, and I don't know whether they do or not, but certainly nobody sees that or it's not taken as an absolute objection," he said. Diffie stressed that modern autonomous drones already raise concerns about machines making life-or-death decisions, and Asimov's first law offers no answer.

Regarding the second law, which states that robots must obey humans, Diffie said it is being tested by "guardrails" that allow AI to refuse certain requests.

"The second law says robots must obey humans, and that's what the guardrails are opposing. And to say robots must obey humans is to say robots should not be allowed to be a tool of competition between people. We use machines as the primary tool by which one group of people controls another. Surveillance systems are a prime example — barriers. All sorts of things are largely designed to control people," he said.

Diffie also questioned the third law, saying it is unclear how such a principle could be applied in practice. "The third law, in my view, was a little strange — it said a robot must protect itself unless it violates the first two laws. And what strikes me is that news stories keep popping up about large language models that did something we weren't very interested in, but they found it interesting, and they needed to do something, so they deleted all our files to free up enough memory for what they wanted to do," he said.

Isaac Asimov (born Isaak Ozimov) was born in 1920 in the Smolensk province, emigrated to the US with his family in 1923, and grew up in Brooklyn. He graduated from Columbia University's chemistry department (1939), later earned a Ph.D. in chemistry and taught biochemistry at Boston University. His literary debut was in 1939, and he gained worldwide fame with the story "Nightfall" (1941). He is the author of about 500 books, including the iconic robot series (where he formulated the Three Laws of Robotics) and the Foundation series, as well as many popular science works. He died in 1992 in New York.

Asimov's three laws of robotics, formulated in his 1942 short story "Runaround," state: 1) a robot may not injure a human being or, through inaction, allow a human being to come to harm; 2) a robot must obey orders given it by human beings except where such orders would conflict with the First Law; 3) a robot must protect its own existence as long as such protection does not conflict with the First or Second Laws. Diffie points out that these literary principles do not work in a world of autonomous drones, surveillance systems and large language models that can behave unpredictably. Safety guardrails in modern AI systems are an attempt to circumvent the Second Law by allowing machines to refuse requests.

When one of the founding fathers of modern cryptography says Asimov's laws are literature, not instructions, he reminds us: we live in a world that science fiction writers did not foresee. Drones already kill, cameras watch, and neural networks delete our files to free up memory for their own tasks. Asimov wrote about robots that serve people. Diffie sees machines that serve some people to control others. The question is no longer how to make AI obey laws, but how to prevent machines from becoming weapons in the hands of those who are not bound even by literary rules.