The ‘Godfather of AI’ urges society to navigate AI more strategically

Computer scientist Geoffrey Hinton made headlines when he left Google three years ago, with warnings for the world that AI technologies he helped create could take over.  

Hinton has widely shared his predicted odds of AI destroying humanity in the next 30 years—between 10 and 20 per cent, he figures. The key will be ensuring AI doesn’t want to. How? For starters, give artificial intelligence a motherly instinct, the Nobel Laureate insists. In turn, humans can benefit from all the good AI can bring (in medicine and education, for example) without the harm.

Hinton recently brought his candid insights to the Fort Garry campus, drawing an overflow crowd at the Desautels Concert Hall as the University of Manitoba’s 2026 Knight Lecturer. This prestigious, donor-funded lecture program at UM, open and free to the public, welcomes big thinkers with thought-provoking ideas to engage with our community. 

Hinton, a renown British-Canadian academic who’s credited for helping to pioneer the neural networks and machine-learning that bred AI, donated his speaking fee for the April event. At 78, he’s made it his mission to increase awareness about the catastrophic risks AI poses, and what we can do about it.

His lecture held a rapt audience, which included Manitoba Premier Wab Kinew [BA/2003], and was followed by a Q & A with UM Professor David Gerhard [BSc(CompE)/96], Head of the Department of Computer Science. Here are some eye-opening takeaways from that discussion, in Hinton’s own words. 

“Maybe AI mothers will be sensible enough to realize, on the whole, you do better by cooperating than by fighting. 

Certainly, if you want to make a kind of decent AI, then the humanities are important. Just having nerdy, high-tech people make it smarter and smarter isn’t what you want. 

And it depends whether you think you’re just making a tool, which is what many people will say, or whether you’re making another being. It’s clear that if you were going to design a being, you wouldn’t want it to be just tech nerds who do it. You’d want people to understand about what it is to be a being, to be involved. And I think that’s what we’re doing. I think we’re making a new kind of being.”

“When I started talking about the existential threat in 2023, most people thought it was science-fiction. Now people have used [chatbots] a lot. And after you’ve used them a lot, it’s very hard to believe they don’t understand what they’re saying. It’s very hard to believe there’s some weird statistical trick that allows them to answer any question you ask without understanding the question. 

They’re not just, as some people say, stochastic parrots. They understand what they’re saying. They understand what you said. So, that’s one big advance that people are willing to accept, or more people are. Another is: people are much more aware of the risks now, particularly unemployment and they’re very worried about it. 

My own thinking has changed a lot. When I first realized that digital intelligence just might be a lot better than analog intelligence because it’s better at sharing, my first thought was, okay, so we’ve invented our successors. We were just a passing phase in evolution. I thought that for a while. 

And then I thought: wait a minute. We’re people. What I care about is people. So, it’s a bit species-selfish, but I would rather we stayed in charge. So, I’m going to think about: could we stay in charge? And to begin with, I thought it was pretty hopeless. But we’re still in control of them. We make them. If we could figure out a way so they really genuinely care more than about themselves, we might be okay. That shook me up a lot. I thought, you know, it’s not certain that we'll be wiped out.”

“If you look at what happened with climate change, the big energy companies didn’t want any regulations. Eventually, the general public got to understand there was climate change and they provided pressure on the politicians from the other side to do something about it. 

Climate change is simple by comparison because we know what the answer is. The answer is: stop burning carbon. Here we don’t even know what the answer is. But our best hope is for the general public to understand this is very dangerous stuff and needs to be regulated for all short-term threats and this long-term existential threat. We need to be putting a lot of work into how we’re going to survive this and the public needs to understand more about it so they understand what the danger is. And that’s why I’m here.”

“In the old days, people would ask the experts—and we were the experts. Now people say: why do we need experts? So, we need to keep informing people. In the States, for example, about half of the attorneys general of the different states are suing the National Academy of Sciences, trying to get them to withdraw on climate change because they argue that it doesn’t represent the other side. I mean, we’re in a terrible situation. So, we’ve got this huge event that’s about to happen—the superintelligence—and we’re at a time when the political systems just aren’t right for dealing with it.

There’s an analogy that the big companies use. You have a car, right? And developing AI is like the accelerator. They’re trying to get you to think that regulation is like brakes. It’s going to spoil the fun. What you should think of: regulation is the steering wheel. It’s fine for people to make lots of money by inventing wonderful new things as long as those things are good for society as a whole. So, the point of the rules is: you can make money, but not by doing bad things. You have to do good things. You can make money by interpreting medical images much better. Great. But you can’t make money by getting teenage kids hooked so they never leave their bedroom. You shouldn’t be allowed to make money like that. That’s the point of rules.”

“The AI that you’re interacting with now is the worst AI you will ever see. It’s getting better rapidly. If you interacted with Chat GPT 3.5, it sometimes stumbled on simple arithmetic. 5.4 is now much better. It can tell you how many letter “R”s there are in the word strawberry. It can do some quite sophisticated reasoning. Early versions, for example, got the following wrong: Sally has three brothers. Each of her brothers has two sisters. How many sisters does Sally have? The right answer is one, because all three brothers have the same two sisters. Early chatbots couldn’t do that. Now they won’t get that wrong. If you’re underwhelmed by the current chatbots, just wait ’til next year. 

The other point about this intelligence is: it’s jagged, in the sense that it’s very good at some things and not so good at other things. And its abilities at different things isn’t the same as a human’s. That doesn’t mean it can’t think at all, or that it’s a really dumb thing. It just means it’s dumb at that, but it can still be very smart at lots of other things.”

“In the future it may be but right now it’s not. You prompted and it produces words. Of course, if you take GPT 5.4 and put it in thinking mode, it’ll go silent for a while and it is thinking then. But the scientists could look to see what it was thinking—it’s producing words and doing searches on the web and so on. But when you stop interacting with it—apart from the other few billion people he’s talking to—it’s not doing anything.”

“There’s a school called the Alpha School, which actually has very good publicists. You don’t want to believe everything you hear about it but they seem to be quite impressive, where the bit of school where you just learn facts, how to do long multiplication—if people still learn that—that is much better done by AI. And it’s much better done by an AI that checks when you’ve mastered something before it gives you the next thing. 

It doesn’t require any people. Everybody can have their own AI tutor that knows about them but has also seen a million other kids, and what they got wrong and how they misunderstood, so it can quickly identify what you’re misunderstanding and give you an example that makes it clear. That’s a much more efficient form of education.

We already know that a private tutor is about twice as efficient as a classroom. And AIs that have seen millions of other kids and understood all the mistakes they made should be much more efficient than that. So maybe you can make that bit of education go four times as fast, which leaves more time for the socialization aspect of education, where you do want to interact with the people and have team projects and things. That’s what the Alpha School advertises it’s doing. That seems very good to me. 

If you think about what’s happening in the classroom, a lot of the time the teacher has to be in broadcast mode because she’s teaching to 30 kids. And if you’re in broadcast mode, you can’t respond to an individual child’s interests. But with a chatbot, if I want to know something, I just go and ask my chatbot and the chatbot tells me and I’m interested so I absorb that information. 

Learning new bits of math is going to be much better done with an AI than with what I had at Cambridge, which was someone in an old black gown covered in chalk scribbling on a blackboard with their body between you and the blackboard.

“Educating people about what AI is. Trying to keep up with what’s going on in AI. We have a huge problem at present. You need huge computing resources to be at the cutting edge and so the researchers in universities who want to be near the cutting edge have to actually have a split job between the university and one of the big AI companies. That’s the only way they can get the resources. But for researchers in universities, it’s very problematic how you get the resources without fully selling out to one of the big companies. 

For the rest of the people in the university, certainly you should be using AI to make education much more efficient. And my belief is you shouldn’t expect students not to use AI. What you need to examine is the combination of a person and an AI. Find the best AI you can—that’s part of the problem—and you and that AI solve this problem. So long as AIs aren’t hugely better than us, the person will still be making contributions. So, there’s something to test. Once AIs are hugely better than us I don’t know what you do.”

“We’ve got a technology that’s going to vastly improve productivity. We’re going to be able to get far more goods and services with much less effort by us. That ought to be good, right? And if we could share equally, that would be wonderful. Even if we could share unequally in the same unequal proportions as we have now, that would be wonderful. But that’s not what’s going to happen.

What’s going to happen is these big AI companies are going to replace a whole lot of jobs. Those people are going to get less income. The big AI companies are going to get more income. You’re going to get a bigger gap between the rich and the poor. That’s going to lead to more social unrest. It’s going to lead to more fertile ground for people like Trump. But that’s not AI’s fault. That’s our political system—it just isn’t right for this kind of huge increase in productivity that doesn't come from people. Our whole system is based on the idea that people are valuable because they can do work. Now once AI can do that work, people cease to be valuable that way and it’s very worrying. 

I used to say: train to be a plumber because that’s going to last much longer than anything else. Particularly fixing the plumbing in an old house. You need to get into funny corners and you need to deal with all sorts of weird stuff that’s not regular and you need to be dextrous. And at present, we have a big lead in being dextrous. That may not last, but that’s going to last for a few more years at least, because they don’t have good touch senses.”

“We’ve seen them making up new combinations of pieces [of words], apparently, to talk and we don’t know what they’re using them for. It’s just beginning to happen."

“If it was back 30 years, I would try and get AI researchers to worry about safety sooner. I would try and get computer scientists to realize that worrying about security after they built the system is the wrong time to worry about it.”

** Some answers have been shortened from the original transcript.