I was watching a news report on TV today about Navi Pillay winning the Nobel Peace Prize. She is 85, and has spent her life as a judge: in South Africa, at the tribunal for Rwanda and at the International Criminal Court. The news reached her in the Nuremberg courtroom where the Nazi leaders were once tried, and her reaction was: "Thank you so much. But being a lawyer and a judge, I still want to verify this information." I really loved that, partly because verification has been on my mind for months. Earlier this year Dr. Werner Vogels, Patrick Johnson and I worked on a talk about trust, Trust is dead: Trust me, which Dr. Vogels gave at the AI for Good Global Summit in Geneva in July. Her one sentence sums up where we ended up.
In the talk we argue that trust has always run on signals that were expensive to fake. In 1488 Amsterdam built the Waag, a public weighing house where merchants' goods were weighed on the city's scales in front of anyone who wanted to watch. Later we trusted institutions, and banks built their lobbies from marble because marble is hard to counterfeit. Then the internet gave us the crowd: a thousand five-star reviews meant a thousand real customers. Every era relied on the same assumption, that the signal cost something to produce.
Generative AI has removed that cost. A face on a video call, an official-looking document and a thousand reviews can now be produced in seconds. In 2024 a finance worker in Hong Kong joined a video call with his CFO and several colleagues, then transferred $25 million. Every one of them was a deepfake. Asking people to look more carefully can't keep up with fakes that are free and endless.
So the talk ends with a fourth era, verifiable trust, where the signal is proof that anyone can check. For those of us who build technology, that means moving from "trust me" to "here is how you can check": systems that verify what goes in, what comes out and what they do. And for all of us, it means borrowing Judge Pillay's instinct. When something feels too good, too urgent or too convenient, verify it.
Conversation