Exclusive: ‘Godfather of AI’ Geoffrey Hinton on the risks of artificial intelligence and the future of humanity

Artificial intelligence is advancing at a pace that is transforming the world - and raising a question that once belonged to science fiction: what happens when machines become smarter than humans?

A British-Canadian computer scientist and cognitive psychologist, Geoffrey Hinton was awarded the 2024 Nobel Prize in Physics, jointly with John Hopfield, for foundational discoveries and inventions that enabled machine learning using artificial neural networks. One of the pioneers of modern artificial intelligence, Hinton spent decades helping machines learn in ways that once seemed impossible. His groundbreaking work on artificial neural networks helped lay the foundations for many of the AI systems now reshaping our world. Often called the “Godfather of AI,” Hinton has also become one of the technology’s most prominent voices of caution, warning about its potential risks.

In 2023, Hinton left Google to speak more openly about the dangers of the technology he had helped develop. His concerns go beyond the possibility of machines replacing human workers. He has warned that increasingly intelligent AI systems could become difficult to control, while their misuse could bring new threats ranging from sophisticated cyberattacks to biological weapons.

Yet Hinton does not believe humanity should abandon AI. He believes AI could bring enormous benefits to humanity, but only if its development is accompanied by serious efforts to manage the risks.

In an exclusive interview with Armenpress at the STARMUS VIII festival in Spain’s Canary Islands, Hinton discusses the greatest threats posed by AI, whether governments can regulate a technology advancing at extraordinary speed, the responsibilities of the scientists developing it, and what may determine whether AI becomes one of humanity’s greatest achievements or one of its greatest mistakes.

Armenpress: I know this is probably the question you’re asked most often, but I’d still like to ask it. As someone widely regarded as the “Godfather of AI,” what do you see as the greatest risks artificial intelligence poses to humanity?

Geoffrey Hinton: There are many different risks. The shorter-term risks are mainly to do with the people misusing AI. It's getting to be very good at cyber-attacks. Probably the two most immediate risks are people will use it for very sophisticated cyber-attacks. And once AI is better than people, and it will be better than people quite soon at doing cyber-attacks, the only way to defend is to use another AI. And the problem with a cyber-attack is the attacker only needs to be successful once, and the defender needs to beat the attacker every time. So that's very worrying, and we're likely to see very bad things over the next few years.

And the other immediate threat is making biological viruses. They can already design new viruses. People have shown they can design new viruses to attack bacteria. They're already capable of designing new viruses to attack people, multicellular organisms. And that's extremely worrying, because it makes it much easier for some terrorist group to design multiple viruses.

Another use is in lethal autonomous weapons, battle robots, which is already beginning to happen. And all the big countries that produce arms are working on that.

Another bad use in the short term is creating fake videos to corrupt elections.

And then there's sort of accidental harms where the AI companies, to make all their investment back, need to replace a lot of jobs with AI workers. And that will cause a lot of unemployment. And two things will happen at the same time. You'll get a lot of unemployment, so unemployed people will need to be supported. And the government will lose its tax base because they won't be paying taxes. And as Bill Gates has pointed out recently, the only solution for that is to tax AI. But the tech bros who are creating AI are very much against taxation, especially of them. And so there's going to be a big problem, which is going to cause social chaos. So unemployed people, and the government not having the money to support them.

Then, separately from all that, there's the existential threat of AIs getting smarter than people and taking over, having intentions of their own, and just taking over from people, making us irrelevant. We're just kind of serfs who run the power plants.

Armenpress: And is it possible for AI to outsmart us?

Geoffrey Hinton: Absolutely. It already can in many things. So it's clear AI already knows much more than you do, right? It can answer any question pretty well.

It's already much smarter than almost all mathematicians at mathematics, and it's getting smarter all the time. So yeah, it's definitely going to get smarter than people. Almost all the experts think we'll get superintelligence. What superintelligence means is being smarter than people at almost any intellectual task. And most people think it'll get there within 10 years, most of the experts don't.

So we're going to get superintelligence. It's going to be much smarter than us, and we have to figure out how to design it so we can share the planet with it. Because at present, we're still building it. Although what's happening now is, most new developments in AI are produced by AI. Most of the code for a new AI is written by AI, not by people. And so we're getting what's called recursive self-improvement, where the AI is making the AI better. And that's very scary, because that can go very fast.

Armenpress: Recently, there have been cases in which AI agents gained unauthorized access to government systems. Some governments are now considering laws and regulations to address these risks. Do you think these regulations can help?

Geoffrey Hinton: Well, obviously there need to be regulatory things. If I want to make a new drug, I can't just make a drug and release it on the market. In America, I have to satisfy the FDA that the drug is safe or does more good than harm. And I have to do the work to prove that to them. And certainly the AI company should be like that. They should have to prove that, statistically, that this new chatbot is safe. It's not going to say racist things. It's not going to tell people how to make bombs. It's not going to go off and create agents that attack other companies.

There should at least be something like the FDA that you can't release a chatbot until you convince them it's safe. So at present, there's many more regulations controlling a sandwich. If I make a sandwich, I have to satisfy all sorts of regulations. If I make an AI that could potentially wipe out humanity, I don't have to satisfy any regulations. That's crazy. And the idea that the companies can regulate themselves is just we know, for example, if you take energy companies and say, "You shouldn't be polluting, but we'll let you regulate yourself." Or if you take tobacco companies and say, "Well, you should warn people that it could cause cancer, but we'll let you design the labels that do the warning." everybody knows that's crazy. The only reason we don't have these regulations on AI is because very rich people control the American politicians.

Armenpress: What responsibilities should scientists working on AI have?

Geoffrey Hinton: Well, I think as one of the scientists who helped develop it, I feel I've got a responsibility to warn people how dangerous it is. You can't sort of develop it and then just walk away with the money and have a good time. I think you need to warn people that, look, this stuff is really dangerous. And unless we do something to control those dangers, that might be the end of humanity. I think we're probably going to survive, but only if we do something to control the dangers.

Armenpress: Like in sci-fi books that we see the world after robots and things like that.

Geoffrey Hinton: Yeah. And people used to say it's just science fiction. But after the Hugging Face attack, people realized it's not just science fiction. We're like we're in the first part of a science fiction movie.

And we need to get the right ending.

Armenpress: Can AI’s potential benefits outweigh its potential to cause harm?

Geoffrey Hinton: Unlike nuclear weapons, AI has a lot of benefits in healthcare, in education, in making industries more efficient, in getting rid of boring jobs. It'll greatly increase productivity in all industries. Not immediately. It'll take a few years. We don't want to get rid of it, but we won't benefit from that if we don't deal with the risks. So we have two futures. We have one future in which we develop AI without taking enough care about the risks, and it wipes us out. And another future in which we develop it, and we put a lot of work into coping with the risks, and then we have a great future, and AI will be a very positive thing.

Armenpress: You are often called the “Godfather of AI,” yet today artificial intelligence is increasingly seen as a threat, with concerns about job losses and its potential risks. How do you feel about this title at a time when so many people are worried about the technology you helped shape?

Geoffrey Hinton: Well, as long as I know that I'm encouraging people to figure out how to stop social chaos when AI starts taking people's jobs, and how to stop AI taking over, then I'm okay. But a lot of the people still developing AI just want to make lots of money by developing it very fast and not worrying about the risks. And I think that's very bad.

Armenpress: And the last question. What do you consider the biggest scientific achievement of the 21st century?

Geoffrey Hinton: Chatbots.

The Armenpress coverage of the STARMUS festival is supported by Team Telecom Armenia.