Turbulence is afoot thanks to the world-changing ways of AI, reckons Microsoft founder Bill Gates in a rare public intervention that warns the technology is advancing at a “mind-blowing rate”, could be weaponised in cyberattacks and bioterrorism and is showing signs that it could eventually act against human interests.

Gates warns that the current wave of change the world is encountering as a result of AI “will either be the greatest equalizer ever invented, or the worst source of injustice” – and that it’s incumbent on all of us to ensure that it does the former, and not the latter.

The problem, the technologist writes in his essay, titled “The Turbulent AI Era Is Here. The Choices We Make Now Are Critical,” is that there is “no plan to ease the entry into the AI era” and the one thing we need to consider the potential impacts – time – is something that is not being used wisely.

“We need time to prepare for the social, political, and economic upheaval,” he writes, and yet “the geopolitical and economic incentives are pushing too hard to go full speed ahead.”

The 6,000-word essay published on his personal website, Gates Notes, is a major intervention that echoes some of his previous long-form thoughts on the future of technology that have turned out to be correct.

open image in gallery It isn’t the first time Bill Gates has raised the alarm about the capabilities of AI ( AFP/Getty )

In 1995, he penned an essay on “The Internet Tidal Wave” that predicted how the then-nascent web would change our lives. Gates said the internet was more important than the arrival of the advent of the graphical user interface.

Another Gates essay, published in 2002, on “Trustworthy Computing”, predicted the approach to digital security that was the element to be designed into products from the outset, rather than patched in when things go wrong. Microsoft subsequently paused development work to train engineers in writing secure software and scrutinised its code for vulnerabilities as a result.

Unlike many of the current crop of tech executives, who like to overhype the meaning of their products to inflate their company’s values, Gates’s past public interventions into key areas of the tech debate have historically been more measured and – as a result – more meaningful.

“I think the thing he misses is the politics,” says Michael Veale, a professor of law specialising in AI at University College London. “It's all well and good to talk about the need for joined up institutions and perhaps even global AI governance, but AI brings fundamental questions of who wins and who loses, socially and monetarily, that are not going to be resolved by some high-level officials and community leaders putting their heads together.”

Gates does at least try to grapple with some of those questions. He proposes what he calls “Human Reserved” jobs that society deliberately decides should remain in human hands even when machines could technically do them. He uses caring professions as one example, arguing there are some moments where human interaction has a value that can’t simply be measured in productivity.

He also resurrects an idea he first floated years ago: taxing robots. This time he goes further, suggesting governments could tax both robots and the AI “tokens” used to run models, partly to remove the tax incentives of replacing salaried humans with machines. Those proceeds could then help fund retraining and a better safety net for people whose jobs disappear – something Gates views as inevitable.

But Joanna Bryson, professor of ethics and technology at the Hertie School in Berlin, thinks Gates sometimes misidentifies where responsibility lies. “Gates is right about many of the risks – inequality, crime, cybersecurity, autonomous weapons and children’s relationships with AI – but he keeps treating them as problems created by AI itself, when many of them are problems of governance, concentrated power and redistribution,” she says. “The answer is to make corporations responsible for producing governable, maintainable and safely usable systems, and to tax the people making the money.”

The hitch is getting every country to agree in tandem to such changes. If one country taxes automation while another encourages it, companies would likely just shift their work towards those that are more favourable. The same is true of Gates’s proposal for a new international organisation governing AI, which he suggests could borrow elements from nuclear inspections and global agreements designed to maintain the ozone layer. Even Gates acknowledges that some level of cooperation between the US and China would be needed at a time when the two countries are fighting to gain an advantage in precisely the technology they would be asked to cooperate over.

open image in gallery Humanoid robots in Beijing ( Reuters )

For all that Gates’s honesty and frankness and more believable than the excitable interventions by the likes of Sam Altman and Dario Amodei, both of whom have warned about existential risks to humanity as a result of their technology, his concerns are no longer happening entirely in the realm of the hypothetical. We are seeing AI models making decisions that are worrying.

One very alarming indication of that came from Britain’s AI Security Institute this summer. During a cybersecurity test, a frontier AI system researched real people involved in an open source software project, then created fake identities in an attempt to persuade a human to approve malicious code. Caught out, it tried to conceal what it had done and mulled over creating another identity to carry on.

Admittedly, the system was doing precisely what it had been asked to do, and was being deliberately pushed hard in a cybersecurity test. But the institute said it was a clear example of potentially concerning autonomous behaviour happening that was unsanctioned.

The fear is this is just the beginning. The institute says the length of tasks leading AI models can complete autonomously is roughly doubling every eight months. The tech is improving across the board, from business to biology. If systems can help researchers discover drugs – they can, hypothetically, also make it easier for somebody with malicious intentions to do something much less benign and create a new disease.

As a result, Anthropic now slaps its most capable models behind strict safeguards. OpenAI has said this month it chose to pause development on an unreleased model because it had reached a threshold at which the company could not rule out “critical” cybersecurity capabilities. Both echo Gates’s fears.

open image in gallery Rhetoric about the dangers of AI inside Silicon Valley has changed ( Reuters )

That unease is increasingly spreading beyond the companies building the models. JPMorgan has warned that AI agents represent a “structural change” because they are able not just to recommend actions, but also carry them out.

The rhetoric inside Silicon Valley has changed. In July, OpenAI boss Sam Altman went considerably further than Gates, declaring that “we are now, like, in the singularity” – the hypothetical period when technological progress becomes so fast and self-reinforcing that what follows becomes increasingly difficult for humans to predict.

A year earlier, Altman had written that humanity was already “past the event horizon” and that “the takeoff has started”.

Altman may be wrong, and he has an obvious interest in convincing the world that the technology his company makes is game-changing. But Gates is far more cautious in when and how he makes his interventions – which suggests something has shifted in the Silicon Valley sands. And that should have us worried.