The man who helped build modern AI now worries he made a mistake
Geoffrey Hinton spent decades teaching machines to think like brains. A new podcast series revisits his story and asks what happens when the technology outgrows its creators.
Key points
- Geoffrey Hinton is a pioneer of "deep learning," the technique at the heart of nearly every modern AI system, from chatbots to image generators.
- Hinton and his collaborators built tools they say they do not fully understand, a gap that worries him deeply.
- The Guardian's podcast series Black Box, first broadcast on 4 March 2024, explores how that gap between creator and creation affects ordinary people.
- Season two of Black Box launches in early September 2025.
Geoffrey Hinton wanted to understand the human brain. He ended up helping invent something that may be harder to understand than the brain itself.
Hinton is one of the founding figures of deep learning, a method of training computers to recognise patterns by processing enormous amounts of data through layers of calculations loosely inspired by how brain cells connect. That method is the engine inside tools billions of people now use every day: the chatbot that answers your customer service query, the filter that catches your spam, the app that labels your photos.
For decades, Hinton and a small circle of researchers pushed this approach when most of the computing world ignored it. Then, almost suddenly, it worked spectacularly well.
So what is the problem?
Neither Hinton nor anyone else can fully explain why it works as well as it does. The systems are trained, not programmed. A human engineer sets the process in motion, but the final behaviour of the model emerges from billions of internal adjustments that no person explicitly chose. Think of it like baking a cake by setting the oven temperature and walking away: you get a cake, but you could not describe every chemical reaction that produced it.
This is the "black box" problem. From the outside, the system produces answers. From the inside, the reasoning is largely invisible, even to the people who built it.
Hinton left his position at Google in 2023, saying he wanted to speak freely about the risks he saw. He has since warned publicly that the technology could cause serious harm if it develops goals humans did not intend and cannot override.
What does this mean for ordinary people?
Most of us will never read a research paper on neural networks. But the systems Hinton helped create already make decisions that touch daily life: which job applications get a first read, which loan requests get flagged, which medical images get a second look.
When those systems make mistakes, the black-box problem matters. It is hard to appeal a decision, or even understand one, when the machine that made it cannot explain its own reasoning.
Guardian journalist Michael Safi spent time tracing Hinton's story for the first season of Black Box, first reported by The Guardian as part of its podcast series Today in Focus. The framing is deliberate: two mysteries meeting each other, the human mind and the artificial one, neither fully legible to the other.
Season two arrives in early September. The first season is available now.
Common questions
Is deep learning the same thing as "AI"?
Deep learning is the most common technique powering today's AI tools, but not the only one. It is the method behind large language models, the technology inside chatbots like ChatGPT, as well as image recognition and voice assistants.
Should I be worried about AI systems I cannot see into?
It is reasonable to ask, when an AI affects a decision about you, whether a human reviewed it and whether you can challenge it. Many regulators are now working on rules that would require exactly that.
