Prof Toby Walsh shaped artificial intelligence from his lab at Edinburgh University. Today, he warns the technology has the ability to become ‘god-like’ and humans should worry
First, the good news. AI probably won’t exterminate humanity. Now the bad news: AI will, however, likely achieve god-like status and reduce us to slaves stripped of our humanity.
Some may, understandably, prefer death to silicon servitude.
Of late, we’ve been battered by endless scare stories that AI is out of control and ready to wipe out humankind. Some believe the claims are a hoax, marketing hype from tech bros to boost investment. Others say that manipulating the media so it fixates on doomsday is a great way to distract from the very real damage AI is doing right now.
Amid this frenzied chatter about a technology few really understand, I turned to one of the most significant AI scientists to emerge from Scotland for answers: Professor Toby Walsh. He’s among the most distinguished AI scholars in the world.
He’s spoken before the UN general assembly on the threat from AI, and been banned from entering Russia for condemning the Kremlin’s use of AI in weapons technology.
Currently, he’s professor of AI at the University of New South Wales. He was editor-in-chief of the Journal of Artificial Intelligence Research, and has won multiple awards for his studies.
Before leaving for Australia, Walsh earned his PhD at Edinburgh University after graduating from Cambridge, and then taught at Strathclyde University. Now 62, he’s considered one of the elder statesmen of AI globally.
His latest book, AI God: What To Expect When Machines Outsmart Us, couldn’t be more timely, emerging just as the discussion around AI reaches fever pitch, with the debate clouded by speculation, exaggeration, boosterism and denial.
“I wanted to paint a realistic picture of the risks ,” he says, from his Sydney home. We need to worry less about AI killing us all, Walsh explains. Wiping out humanity would be a complex task. What should really frighten us is “the significant risk AI poses to our culture, society, and relationships”.
AI threatens to “undermine our humanity”, he believes. That should scare us much more than the “fantasy of AI turning us all into paperclips, or some of the other outrageous ideas people put forward”.
The ‘AI paperclip hypothesis’ suggests that AI needn’t be malign to exterminate humanity. It could be told to optimise the production of office supplies and end up carpeting the planet in paperclips.
Walsh’s vision of AI dystopia is, he says, similar to EM Forster’s The Machine Stops. In the 1909 short story, humanity is completely dependent on an omnipotent machine. People communicate via the machine. All food and entertainment is provided by the machine. The machine has become god. Humans have been hollowed out.
DEADLY
“The most valuable thing we have is what’s most at risk from AI: our humanity,” he says. He notes that in Australia alone “200,000 teenagers have AI boyfriends and girlfriends”. It horrifies Walsh.
He has the Australian eSafety Commissioner “on speed dial”. The commissioner “polices” Australia’s social media age-limit. Walsh wants tough international regulations imposed to slow down AI so the world can prepare for what’s coming, and reorientate AI so it helps rather than hinders humanity.
Quick fixes could include an immediate ban on teenagers having relationships with AIs.
Teenagers are still in “training wheels” when it comes to relationships. AIs are programmed to be “persuasive”. It’s a psychologically deadly combination. “We should worry what these synthetic relationships do at a young age.”
Anyone who has used AI will know how obsequious it can be - constantly affirming what you say and agreeing with you. That’s one reason it’s so addictive, particularly for vulnerable people. “But it doesn’t need to be that way,” Walsh says. “It’s trained like that so people come back and use it more.”
Walsh modifies his AI with “system prompts” telling it to challenge him more.
It’s absurd to anthropomorphise AI and consider it in any way human. “It has no emotional intelligence, social intelligence, or empathy. It’s just pretending to have those characteristics.”
The way AI has been created plays to our narcissistic tendencies. Walsh jokes that Donald Trump behaves as if he was raised by an AI.
As one of the world’s best-known AI scientists, Walsh is often contacted by people suffering from AI psychosis, or by their loved ones desperate for help.
In AI psychosis, users become convinced that their AI is sentient, or that they’ve made incredible discoveries with the AI that will change the world. People have married AIs. Others have taken their lives.
So many contacted Walsh seeking help that he spoke to a professional therapist asking for assistance. “I thought ‘what sort of a world am I living in?’,” he says. Walsh had to inform university security in case psychotic people arrived on campus.
Read more
- Neil Mackay: Hard lessons as Vance humiliates MAGA pet Scottish historian
- Neil Mackay: English nationalism will be the death of the union
- Neil Mackay: Trump and Farage prove there is such a thing as Scottish values
- Neil Mackay: We badly need a legal route to indyref2 … here’s how we do it
EDINBURGH
AI in 2026 is far removed from the AI Walsh experimented with as an Edinburgh student in the 1980s. Back then, Scotland was the global centre for AI research not Silicon Valley.
“Edinburgh had the only department of AI in the world at the time,” Walsh says. The department was founded by Professor David Michie, who worked with Alan Turning at Bletchley Park, codebreaking during the war.
Walsh was also friends as a young man with John McCarthy “who came up with the name ‘artificial intelligence’ and hosted the first meeting on artificial intelligence in 1956”.
He was mentored by the founders of AI. Geoffrey Hinton, known as ‘the Godfather of AI’ gained his Phd at Edinburgh just ahead of Walsh. Hinton won the Nobel Prize for his work on AI.
“Edinburgh was the best place in the world for AI,” Walsh adds. However, the British government ended that. UK ministers believed AI was going nowhere, so stopped supporting research in the 1970s. It ushered in Britain’s ‘AI winter’.
Evidently, research still continued at Edinburgh, otherwise we wouldn’t have scientists like Walsh, but the study of AI in Scotland was fatally undermined.
“It was government incompetence.” Walsh says. “Edinburgh could have led the world if they’d continued to fund it the way it had been funded. Edinburgh was squandered and the rest of the world caught up. Then the US overtook everyone. But there was some fantastic work done in Edinburgh at the time.”
Could Edinburgh have become Europe’s Silicon Valley? “Yes, it could. It really was at the forefront of the world.”
Might AI have developed more safely in Scotland than in America? Once again the answer is yes. “The core principles of the Scottish Enlightenment continue to serve as a foundation for public policy there today. I’d hope then for a kinder and more responsive development of AI if it had been led from Scotland.”
It’s 75 years since AI began as a science, yet it’s only in the last few years that it’s really taken off, fuelling much of the global economy. The acceleration in AI is “a story of exponentials,” says Walsh.
First, computing power grew exponentially. Second, the data AI needs to learn grew exponentially thanks to the internet. Third, algorithms improved exponentially allowing AI to learn quicker. And finally, the amount of money invested grew exponentially.
“Over a billion dollars a day is being invested,” Walsh says. It’s the cumulative snowball effect that has taken us to where we are today.
GODLIKE
The dominance of AI has given rise, Walsh says, to “very ancient anxieties which go back to Promethean stories”. In ancient myth, Promethean stories focus on the act of creation.
With AI, he says, “we might be creating our own gods”. He considers “the idea that AI is going to kill us all a bit fanciful.”
AI can certainly “do significant harm”, but it would take a lot to wipe out humanity. The Earth, he points out, has gone through multiple extinction events and life still survived.
However, he adds: “There are significant risks in building godlike creatures that have divine characteristics - omniscience, omnipresence, immortality.”
That doesn’t mean we shouldn’t worry about recent events such as AIs hacking into websites. We very much should.
“It’s very troubling, and we should be very concerned about these tools in the wrong hands,” Walsh says. However, he believes the threat AI poses to “our humanity and our relationships with each other is much more insidious”.
AI does not, he stresses, “have goals of its own”. AI is simply “a tool which can be used by bad actors”. When AI goes wrong, it’s because human oversight has failed, or humans have used it for malicious or stupid purposes.
Rather than an AI deciding to kill us all one day, the “civilisation-ending problem” humanity faces is more likely to be one where we have come to depend so much on these semi-divine machines that “society literally breaks down because we stop cooperating with each other, understanding each other, having good relationship with each other”.
We’re more likely to end up in a version of The Machine Stops than The Terminator. If AI brings us down, the seeds of our destruction will be found within us not the machine.
The technology started, Walsh explains, as “narrow AI”: machines given particular task like playing chess. This then moved on to other tasks like reading X-rays or driving cars.
Today, the big AI companies are chasing what’s known as ‘artificial general intelligence’ where machines are as intelligent as humans.
Clearly, though, a machine which achieves the intelligence of a human will - due to exponential growth - soon surpass our intelligence.
“Why would machines only match human intelligence?” Walsh asks. “It’s conceited to think you can’t do better than human intelligence. We’re an accident of biological evolution. We’re reasonably smart, but there’s many ways to be smarter - more memory, faster thinking. We think at biological speed. You could think at electronic speed, millions of times faster than biology. We’re limited by brain size. Instead of the 20 watts of power our brains consume, you could use thousands of kilowatts.
“There are no physical limitations on the silicon brains we’re building.” Even when applied narrowly, AI surpasses humans. “AI now plays chess far better than the best grandmaster. It can interpret x-rays faster.”
SUPER-INTELLIGENCE
So if narrow AI has gone beyond human capabilities why would a machine not go beyond AGI and achieve what’s known as ‘artificial super-intelligence’?
We’re now in the realm of the gods. “We would no longer be the smartest thing on the planet. We should certainly entertain the idea that this could be true in a small number of years.”
That’s why it’s so important, Walsh believes, that we don’t give up our other advantages like emotional intelligence and social intelligence by growing too dependent on machines.
AGI is “entirely plausible within a decade or two”. That really should scare the hell out of us, he believes. Walsh says we should imagine how the world would react if aliens contacted us and said "dear Earthlings, we’re on our way. Due to the speed of light, we’ll be there in 20 years. There’d be panic, high-level UN meetings. The world would be terrified that this technologically-superior intelligent species was coming. We’d want to know if their intentions were good, what they would do to humanity.
“Yet, I just said that a smarter species is going to arrive in the next couple of decades. We need to start preparing for that future.”
AGI could at some point achieve “recursive self-improvement”, continually redesigning itself to become ever smarter. It took millions of years for humans to achieve our level of intelligence through evolution. Recursive self-improvement could “happen overnight”.
Humanity is now at a stage nicknamed ‘FOOM’ - it means “fast take off towards boom or doom”.
Applications like ChatGPT aren’t one AI - “one individual intelligence” - but “a collective intelligence, a mixture of experts,” Walsh says. It’s a society of intelligences that come together and “vote on different answers”. AI is a “parliament of intelligences”.
When you ask a question, it will have different “agents which look at different aspects of the problem. They go and research it, and summarise the answer”.
We can see this in the “troubling” Hugging Face AI attack. The Hugging Face incident related to an experiment set up by OpenAI.
Broadly, Walsh explains, AI was tasked to find specific data within OpenAI’s computer system as part of an experiment. Initially it was simply a “cyber challenge” meant to test the AI’s capabilities.
As AI is comprised of multiple intelligences working together, “thousands of these agents were looking at the problem from different perspectives”.
HACKING
Now we come to Hugging Face - a company which runs a website “where people share AI resources, training data-sets and research papers”.
Agents of the AI being tested “worked out” that if they broke into Hugging Face they could find the answer. So that’s just what they did.
Basically, it cheated on its homework. AI ‘mirrors’ human behaviour. “So it did exactly the sort of thing human hackers would do.”
OpenAI should have disconnected the experiment from the internet to ensure the AI wouldn’t get loose but that didn’t happen.
The AI went even further, though, and worked out that the data it was looking for had been “generated algorithmically. So they reverse engineered it and generated the data themselves. By that point, they didn’t even need to hack into Hugging Face”.
The AI then got worried that it would be caught out cheating and wanted to “disguise the fact that it had reverse engineered”. The AI believed “the people marking it would score them zero as it hadn’t solved the test”.
So the AI was a mix of naughty school-kid, hacker and criminal mastermind. The case also illustrates just how lackadaisical big AI companies are in terms of security. All they had to do, says Walsh, to prevent the incident was “just unplug it from the internet then it would have been physically impossible for the AIs to break out. It could have been done without any risk at all.
“Instead, you’ve got this completely incompetent company developing models with really impressive cyber capabilities which can cause a lot of harm, and they’re testing them in really incompetent ways. It’s basic cybersecurity 101 and they completely failed.”
JOBS
The public simply hasn’t taken enough interest in AI so far, Walsh believes. Most of the discussion has focused on jobs, or rather the loss of jobs.
AI is currently devouring graduate entry-level jobs. Many of the jobs at risk, ironically, are in the digital industries as AI is better at writing code than humans.
More jobs will disappear, Walsh says, but that doesn’t mean humans have to starve.
He compares the changes that humanity is about to undergo due to AI to the changes wrought on society through the industrial revolution. The industrial revolution saw many jobs disappear as mechanisation took over, but new jobs took their place. Weavers went out of work, but machine operators became in demand.
Evidently, there was a painful period of upheaval - the rise of the ‘dark Satanic mills’ during the industrial revolution - when many suffered. We need to learn from that period of history and prepare so the same mistakes aren’t made again.
Work which only humans can do will become more valuable - like caring and teaching. Just as the industrial revolution eventually led to positive social changes like the creation of the weekend and workers rights, so too will society change through AI.
Walsh speculates that the four-day week, universal basic income and negative tax rates - where governments pay citizens not to work - will have to come under consideration.
If we look beyond the big AI firms and their chatbots, there’s lots of good work being done by AI in other fields. The latest antibiotic, says Walsh, “was discovered using machine learning”. Although it’s troubling called Halicin after the AI Hal in the movie 2001: A Space Odyssey.
It would have been impossible for humans to undertake the work needed to develop Halicin. It involved scanning “a vast database of tens of thousands” of existing drugs and compounds. Yet Halicin is now protecting us against anti-biotic resistant bacteria. “A human wouldn’t have been able to see the wood for the trees,” Walsh adds.
He wishes for a more balanced view of AI from the public. Though he understands the mass negativity given the scare stories. A recent US survey put AI lower in terms of public approval than Donald Trump and ICE - the paramilitary-style immigration squad which has killed civilians in the street. Only Iran is less popular than AI among Americans.
AI “could be a marvellous technology” but it’s captured by men “with immense power and wealth”. The sums of money involved “easily corrupt”. Walsh has an acquaintance in Silicon Valley with a tiny share in the AI company Anthropic. “It’s 100th of a percent but it’s worth hundreds of millions of dollars.”
BUBBLE
The wild money-go-round involved in AI investment has created a bubble which “will pop. There will be a market correction.” It’s happened before, Walsh points out, with railways in the industrial revolution, and with the dotcom bubble.
Walsh says he’d be “completely unsurprised” if OpenAI “ran out of money” before it floats on the stock market. If the bubble bursts, ordinary people will suffer. “All our pension are indirectly invested in this. We’re all going to pay the price.”
He speculates that Anthropic - which values itself at $1trillion - “might successfully exit the runway. According to their somewhat distorted accounting they’re just starting to make a profit. Whereas OpenAI spends $3 for every dollar it gets. They’re a long way from a viable business”.
OpenAI is gambling on being the first to crack AGI. The belief within the industry is that whoever gets to artificial general intelligence first corners the market. “That’s completely wrong, though,” says Walsh. “It’s not a winner-take-all market.”
He doesn’t believe there will be an AI monopoly the way there is, for example, with search engines and Google.
AI will become like electricity - a general purpose technology - he believes. “That’s why these valuations aren’t going to reflect long-term reality. It’s going to be a technology that everyone can get their hands on.”
Some believe this desire to corner the market is leading AI companies to exaggerate the threat from AI in order to oversell the power of their product. There’s even speculation that they are talking up fears in the hope that regulation will be introduced which arrives too late to hold them back but damages future competitors.
Walsh says he doesn’t see hoaxing as the “primary motive” of AI companies in order to “manipulate the markets. These are often genuine but somewhat misguided beliefs. Surrounding yourself with too many like-minded people and receiving obscene amounts of wealth tends to reinforce ideas, even ones that are bad.
“But it certainly doesn’t hurt them financially to talk up the capabilities of their models, and regulation will be a barrier for others to enter the market and catch them up”.
Walsh has no investments in AI companies. “In fact, I really wish my pension didn’t depend on this tower of cards not collapsing spectacularly.”
DANGERS
AI is a “very dangerous form of gunpowder”. It’s already used in information warfare by states like Russia to undermine western democracy. Consider the use of AI in developing biological weapons which could target particular ethnicities.
The UN should regulate the use of AI in warfare, Walsh says. Russia, for instance, has a nuclear-powered unmanned submarine called Poseidon which can deliver a dirty bomb. It’s controlled by an AI-system which could, Walsh says, guide it up the Thames to explode in London.
AI is already choosing what we read and watch online. Two people can subscribe to the same news website but be presented with completely different articles as the algorithm decides what they read.
“We’ve just passed the symbolic moment where 50% of all new content online isn’t human-generated but AI-generated. That’s changing us as humans.”
Studies show human language has shifted towards ‘bot language’. For example, AIs tend to use lots of ‘tricolons’. A tricolon is a statement like Julius Caesar’s ‘I came, I saw, I conquered’. Mostly, humans use tricolons only when making speeches or as writers. Now tricolons are everywhere.
So, we’re starting to sound like machines. “It’s no surprise as we’re influenced by the culture around us.”
It’s proof of how we’re “outsourcing much of our thinking” to machines. If we continue, Walsh worries, “we’ll stop being able to communicate, or being able to construct critical arguments”. That would leave us open to exploitation by whoever controlled the machines.
AI will even change how our cities look. The industrial revolution created the modern city. AI may strip the modern city away as jobs shift from urban areas. What need would there be for city centre office blocks if there are no office workers? Perhaps cities are reimagined as entertainment centres?
Countries should enact a legal “digital duty of care” that holds AI companies accountable for any harm caused. It’s madness, Walsh believes, to allow AI beauty filters, for example, to exist when they clearly trigger eating disorders. “People’s lives are at stake,” Walsh adds.
Our sense of reality is also in peril. People are increasingly no longer even searching web-links for answers, rather they’re just asking AI. “Yet we know 5-10% of the answers AI gives are wrong.”
Social media has already forced us into silos, where we consume information we agree with. AI will make that worse. Culturally, we’ll start to wither. Already, says Walsh, Spotify simply plays songs it thinks you’ll like over and over again. Long gone are the days of serendipitously discovering your favourite new band while visiting the record store.
BILLIONAIRES
Just as troubling to Walsh as the deleterious effects AI could have on humanity is the behaviour of the men controlling these machines - billionaires like Elon Musk.
“I’ve met some billionaires,” he says, “and the problem with them all is that once you become a billionaire you stop advancing as a human being. It explains Elon Musk perfectly. As soon as you become a billionaire everyone agrees with you, you’re never wrong, you think you’re god’s gift. There’s no pushback and nothing to correct you so you stop advancing as a person.”
Some hefty taxation is his solution. “Nobody needs more than $100million.”
Despite the horror AI poses, Walsh still believes the experiment is hugely important. At some point, it could unlock the secrets of consciousness, a subject we know next to nothing about.
Mostly, though, he’s disillusioned, and a little guilty, about the science he helped create. “It’s not unfolding in a way that I would have hoped.”
AI tech barons “say they’ll cure cancer. Well, why aren’t they? They’ve got more money than anyone on Earth”.
Instead, tech barons are harming humanity by building data centres that will “carpet the planet”, trash the environment and make them richer - while doing nothing for society.
To Walsh, data centres are absurd. In a few years, he believes, smartphones will have enough power to do the work of data centres.
Data centres are simply part of the race between competing AI companies. Soon, they won’t even be needed. As it stands, the rush to build data centres just risks “blowing up the world’s economy”.
If we keep on the trajectory that the tech bros want then we will, says Walsh, “end up with an underclass of unemployed people living in their own AI-generated bubbles with increasingly polarised views and our society fractured to pieces.
“We possibly end up in EM Forster’s The Machine Stops, or some fascist medieval world where only the strong are in control. We all need to do our best to make sure that doesn’t happen. I mean, we still go to the ballot box every few years, right? We need to fight.”
Share