Tech CEOs are locked in an artificial intelligence "arms race" that risks wiping out humanity, top computer science researcher Stuart Russell told AFP on Tuesday, calling for governments to pull the brakes.
Russell, a professor at the University of California, Berkeley, said the heads of the world's biggest AI companies understand the dangers posed by super-intelligent systems that could one day overpower humans.
To him, the onus to save the species rests on world leaders who can take collective action.
"For governments to allow private entities to essentially play Russian roulette with every human being on earth is, in my view, a total dereliction of duty," said Russell, a prominent voice on AI safety.
Countries and companies are spending hundreds of billions of dollars on building energy-hungry data centers to train and run generative AI tools.
The rapidly developing technology promises benefits such as drug discovery, but could also lead to job losses, and facilitate surveillance and online abuse among other threats.
Alongside that is the risk of "AI systems themselves taking control and human civilization being collateral damage in that process", Russell said in an interview at the AI Impact Summit in New Delhi.
"Each of the CEOs of the main AI companies, I believe, wants to disarm" but cannot do so "unilaterally" as they would be fired by investors, he said.
"Some of them have said it in public and some of the told me it privately," he added, noting that even Sam Altman, head of ChatGPT maker OpenAI, has said on-record that AI could lead to human extinction.
OpenAI and rival U.S. startup Anthropic have seen public resignations of staff who have spoken out about their ethical concerns.
Anthropic also warned last week that its latest chatbot models could be nudged towards "knowingly supporting -- in small ways -- efforts toward chemical weapon development and other heinous crimes".
International gatherings such as this week's AI summit provide an opportunity for regulation, although its three previous editions have only resulted in voluntary agreements from tech companies.
"It really helps if each of the governments understand this issue. And so that's why I'm here," Russell said.
India is hoping the five-day AI summit, attended by tech bosses and dozens of high-level national delegations, will help it power ahead in the sector.
Indian IT minister Ashwini Vaishnaw said Tuesday that the country expects more than $200 billion in AI investments over the next two years, including roughly $90 billion already committed.
Meanwhile fears that AI assistant tools could lead to mass redundancies in India's larges customer service and tech support sectors has caused shares in the country's outsourcing firms to plunge in recent days.
These kind of back-end jobs in India are ripe for replacement with AI, Russell said.
"We are creating human imitators. And so of course, the natural application for that type of system is replacing humans."
Russell is sensing a burgeoning backlash against AI, "particularly among younger people".
"They actually are pushing back against the dehumanizing aspects of AI," he said. "When you're taking over all cognitive functions -- the ability to answer a question, to make a decision, to make a plan... you are turning someone into less than a human being. The young people do not want that."
© 2026 AFP
13 Comments
Login to comment
SomeWeeb
It's always something that we are arms racing over.
Ricky Kaminski13
We are all hoping that Terminator and the Matrix were just cool SF movies and not a window into our inevitable future enslavement, demise or even extinction.
I believe in humans and am a AI optimist, but that may turn out to be naive as some of the top players in the industry give humanity a 50/50 chance of survival. Some even less! Remember SNS was supposed to bring is all together and inform! How’s that working out? Feeling the love?
AI is gunna exponentially more of a game changer too. We could be living as the last crop of pre-cyborg real humans. What a time to be alive ay rockers? We are the generation that spawned a new species of super intelligence, one that will surpass our own in the very near future. How it will treat its parents remains to be seen. Personally I like to say please and thankyou to my AI chatbot, just for good measure!
virusrex
There is no need for theoretical scenarios to argue for the need of controlling AI deployment, it is already causing serious problems (like disinformation, sexual harassment, etc.) and the energy requirements make much more difficult to correct climate change.
Then again we also have people endlessly repeating that AI is impossible because of some "mathematical reasons" that nobody can produce, explain or see.
Sven Asai
Maybe I am the addressed, lol Yes, there are a lot of mathematical reasons, and of course some others, too. You say already yourself, what they lead to. And of course everyone can produce , explain and see them. Multiply matrices, check the several trigger functions, try to project irrational numbers and number space densities into the digital limitations of using only ones and zeros, and all such issues and more. No need to like my viewpoints, but we are already common in the sense, that it is not working and causes serious problems, aren't we?
virusrex
Actually no, there are no mathematical reasons, you promised to bring those reasons since november last year, and yet you could not present anything.
Nobody can, including you, because they don't exist. The difference with everyone else is that only you are claiming they exist, but can never produce any.
TaiwanIsNotChina
So does social media, but we had no qualms about unleashing that on the world.
1glenn
Not a scientist, but I do not see how we survive the AI revolution. How are we supposed to control AI development? Seems to me that once machines can design and build other machines without humans in the loop, it is game over......unless machines choose to keep us as pets.
Wesley
Go after the AI companies and the AI programmers. Problem solved.
It's not the AI itself that's the problem; it's those who own the companies and those who program/code the AI that are the problem.
Sven Asai
OK, wait and see. It is all destined to fail, because mathematically impossible. And of course I have published here and there some of the reasons, although JT and other sites are not the usual target to publish them. It is not my fault, that we cannot discuss here in detail and in a scientific way. This is a news platform on Japanese topics, not a university intranet on AI. I gave a lot of clues and that's it. I am not even paid for explaining and throwing pearls. So just believe it or taste the high costs of failing big.
ian
Once AI becomes self aware or true AI. But that's probably the same point they can design and build on their own.
Still, they need energy, if they can produce/gather enough energy by themselves then that's the point they can actually turn successfully against humans.
PavlovaLover
"It is all destined to fail, because mathematically impossible"
I'm not asking you for a university-level course in AI. I am asking you for links to information that support your contention. Thus far, and going back for months, you have not provided any information that actually does support your "mathematically impossible" narrative. On any other topic I would be forced to accept that you are inventing a problem where none exists.
ian
Impossible depends on what you are trying to achieve.
What is mathematically impossible in this case?
It's destined to fail in what way?
Sven Asai
There is not such a thing like an external link to my own research and public statements other than you can read here or on related sites I visit. If you mean other published and more trusted opinions , maybe you find them with a specific search. A good source for example is Medium Daily Digest, where they cover many pro and cons on anything AI, Math and other near topics.
A real AI will fail, one that deserves the name and works without so many errors and is very or sufficiently reliable. It doesn't mean that the many models don't work, but at minimum 70% are producing total garbage, only the very few best models out of millions, make it into the headlines. And those still have errors, produce fake, have halluzinations or deny answers at all etc, which is caused by the math limits I have in focus and gave hints about.
So mathematically impossible mainly means, it is not possible to project the natural intelligence, like in neurons or whole brains for example, into the digital computer world, which is finally restricted and limited to voltage level based zero and ones in the circuits or memory cells, by mathematical methods like they are used in AI modelling, like multiplication of matrices, a catalogue of trigger functions, LLM and other model's backward stepping and the like.