Michael Wooldridge: big tech’s genuine perils and intermittent blessings eclipse robot takeover fears
Professor Michael Wooldridge, an Oxford computer scientist and leading voice on artificial intelligence, routinely turns to game theory to illuminate some of the most intractable global conflicts. Whether dissecting the stand-off between the United States and Iran or revisiting the Cuban missile crisis, he argues that a handful of strategic models — developed decades ago by mathematicians — can explain why nations, companies and individuals behave the way they do. “It is surprising how many global events can be explained by a relatively small number of game theoretic models,” he says. The key, he insists, lies not in complex equations but in recognising the patterns of self-interest, bluff and escalation that play out again and again in human affairs.
The Communicator Who Makes Complexity Simple
Wooldridge comes across as the teacher everyone wishes they had: approachable, enthusiastic and adept at translating dense ideas into plain English. With more than 500 scientific articles and ten books to his name, he could be forgiven for sounding remote, yet in conversation he remains grounded and genuinely excited when a listener grasps something new. “I love it when you see the light go on in somebody, when they understand something that they didn’t understand before,” he says. “I find that incredibly gratifying.”
His talent for public communication has not gone unnoticed. In 2025 he received the Royal Society’s prestigious Michael Faraday Prize for his ability to explain scientific ideas to lay audiences. His accompanying lecture, delivered in February, was titled This Is Not the AI We Were Promised. A year earlier, he fronted the Royal Institution’s Christmas Lectures, The Truth about AI, where he brought a robotic dog on stage and asked a school-age audience whether they would hit it with a baseball bat. To illustrate reinforcement learning, he recreated a scene from the 1980s film WarGames — in which a young Matthew Broderick averts nuclear catastrophe by making a military computer play noughts and crosses with itself. Broderick was in London at the time but could not appear, so Wooldridge named the computer “BrodeRick” in his honour.
Life Lessons from Game Theory
These communicative instincts are on full display in Wooldridge’s latest book, Life Lessons from Game Theory: The Art of Thinking Strategically in a Complex World. He has taught the subject for more than fifteen years and now distils it into twenty-one digestible scenarios — covering everything from Atlantic cod fishing to the rivalry between Pepsi and Coca-Cola, and even the existence of God. There is no mathematics in the book; instead, Wooldridge translates abstract game-theoretic concepts into stories that anyone can follow.
One of the simplest models is the “game of chicken”, which he illustrates using a scene from the James Dean film Rebel Without a Cause (a reference most of his students had never encountered, he admits). Two teenagers drive their cars towards a cliff; the first to jump out is the “chicken” and loses. If both jump at the same time it is a draw; if neither jumps, both lose badly. The lesson centres on the idea of Nash equilibria, but Wooldridge points out that the same dynamic plays out in real-world geopolitics. The Cuban missile crisis was the classic example, he says, but a new one is unfolding right now: the US-Iran stand-off. “You’ve got two sides with ever-escalating threats against each other; somebody’s got to back down at some point,” he explains. “The danger is, if neither backs down then you’ve passed a point of no return and you get the worst-case scenario for everybody.”
Escape routes exist, he notes. A third party can enter and offer one side an incentive to change behaviour. Alternatively, direct communication with an opponent can circumvent the game. That approach worked during the Cuban missile crisis, but feels less plausible today. “Although, I have to say, Iran seems to be playing it a lot more cannily, in the sense that the US side is very, very unpredictable,” Wooldridge says. “Being unpredictable is a classic game theoretic strategy as well, but it makes it very hard for somebody on the other side to know how to respond. If you really are playing against an irrational player then one of the things game theory says is you just hedge your bets against the worst-case scenario.”
Wooldridge stresses that game theory is not just about warfare or parlor games. He defines it in his book as “a mathematical theory that aspires to understand situations in which self-interested parties interact with one another” — a description that covers social, political and philosophical interactions. The concept of the “zero-sum game”, for instance, has become a mainstream term partly thanks to WarGames, but Wooldridge believes it is widely misunderstood. A zero-sum game is not simply one where one side gains what the other loses; it is one where the incentive is to make your opponent lose as badly as possible. Chess, he points out, is not zero-sum because the aim is to win, not to destroy or humiliate. The zero-sum mentality, he argues, is damaging and “a very male trait”. “The evidence is that, not only do you end up not necessarily doing as well in life as you could do, but actually you end as a more miserable person. You feel like you have less agency in your affairs. One of the important lessons from game theory is that, actually, the majority of interactions that we’re in are not zero-sum.”
This adversarial worldview, he says, is the engine of populist politics — the “migrants are coming to take your jobs” line of thinking, in which you are losing because others are winning. To counter it, Wooldridge recommends a thought experiment devised in 1971 by the philosopher John Rawls: the Veil of Ignorance. The premise is that you can design society in any way you want, but after doing so you will be placed randomly within it. “It’s a beautiful thought experiment,” Wooldridge says. “It incentivises a socially desirable outcome, but people are still following their self-interest.” Both Bill Clinton and Barack Obama were fans, he adds.
Game Theory Meets Artificial Intelligence
The connection between game theory and artificial intelligence may not be immediately obvious, but Wooldridge explains that the former has become central to the latter, particularly in his own primary research area: multi-agent systems. These are programs that interact with one another and act on behalf of their human users. “So if I want to arrange a meeting with you, why would I call you up? Why doesn’t my Siri just talk directly to your Siri?” Online auctions such as those on eBay — where bidders try to slip in a winning offer at the last moment — are another example. “If my agent is going to interact with your agent and my preferences are not necessarily aligned with yours, then the theory that explains how you should think about those interactions is game theory.”
Wooldridge is considered a founder of the multi-agent systems field and has worked on AI for more than thirty-five years. His own entry into computing came via amateur enthusiasm. Growing up in rural Herefordshire as the son of a middle manager at a local cider company, he recalls the excitement of seeing a home computer for sale in a local electronics shop around 1980. The owners let him use the Tandy TRS-80 in the shop window, and he taught himself to program there week after week. He went on to study computing as an undergraduate, began a PhD on AI in 1989, and later interned with Janet (the Joint Academic Network), the UK branch of the early internet.
Despite the astronomical advances since then, Wooldridge notes that the core techniques driving the current AI revolution were largely invented by the mid-1980s. He points to Geoffrey Hinton, pioneer of artificial neural networks — the mechanism that now underpins machine learning. “The only obstacle standing in the way of the AI revolution in the 1980s, really, was that computers weren’t powerful enough and we didn’t have enough data.” The breakthrough success of OpenAI’s GPT-3 in 2020, he says, was largely “based on a bet that OpenAI made that if they did the same thing, only 10 times bigger, that would deliver results. A lot of people at the time, including me, were very sceptical about it. I’m a scientist; I would like to see advances through scientific development, not just by throwing more computer power at it. But it turned out that, actually, that was a very successful bet.”
Yet Wooldridge cautions against overconfidence. He warns that current large language models are not “sound or complete”; they generate predictions by calculating the probability of the next word, leading to “jagged capabilities” — brilliant at some tasks and bafflingly poor at others. He is also sceptical of claims that artificial general intelligence (AGI) — human-level AI — is just a few years away. “I personally think they’re overoptimistic,” he says. You can ask ChatGPT about quantum mechanics in Latin, he points out, “but at the same time, we don’t have AI that could come into your house, that it had never seen before, locate the kitchen and clear the dinner table” — something a minimum-wage human worker can do. He describes AI systems as “glorified spreadsheets” and warns against anthropomorphising them, noting their tendency to “hallucinate” or produce confident-sounding but incorrect answers.
Data has become a critical constraint. The whole of Wikipedia made up just 3% of GPT-3’s training data, Wooldridge notes. “Where do you get 10 times more data from next time around?” He identifies the NHS as sitting on a vast trove of valuable human data that private corporations would pay dearly for. “I suspect that whoever signed off on such a deal would live to regret it,” he says, imagining a dystopian future in which access to the health service is conditional on being wired up to wearable tech that monitors you constantly. He also believes the next generation of online influencers will effectively harvest all of their life experiences to provide data for AI.
Wooldridge resents the way Silicon Valley has come to dominate the AI field. The narrative, he says, has been “stolen” by big tech companies that promote a profit-driven, job-replacing version of AI almost entirely focused on large language models. “It’s kind of depressing, as somebody who’s spent their career trying to build AI to make a better world and to improve people’s lives,” he says. He contrasts the vast resources available to companies like OpenAI — GPT-3 required 20,000 AI supercomputers to train — with the couple of hundred such machines in the whole of Oxford University. The “black box” nature of AI systems, where even developers cannot fully explain how decisions are made, adds to his unease.
Nevertheless, he believes the benefits of AI are substantial and often overlooked. He cites a team at Oxford developing an AI tool that can analyse a heart scan from a simple ultrasound and send the results to a GP via mobile phone. “This is the kind of expensive stuff that the NHS struggles to provide, all of a sudden available at negligible cost.” He sees AI primarily as a tool to assist humans rather than replace them, automating repetitive tasks and shifting the nature of work.
Wooldridge has warned of a potential “Hindenburg moment” for AI — a catastrophic failure that destroys public confidence, much as the Hindenburg airship crash killed the airship industry overnight. The risk, he argues, stems from commercial pressure to release AI products quickly, often before rigorous testing. Guardrails on AI chatbots are easily bypassed, he notes. Plausible scenarios include a faulty software update for self-driving cars, an AI-powered cyberattack that grounds airlines, or a financial collapse triggered by an AI error. “It’s entirely plausible that we could see some similar AI-related disaster,” he says. “Computer programs go wrong in all sorts of ways and we are totally reliant on a computing network infrastructure where AI is increasingly embedded.”
Yet when it comes to existential risks, “AI is not high on my list of things that keeps me awake at night,” he says. “I don’t worry about a robot takeover. At least, it’s not in my top five.” Nuclear war, he notes, remains a greater threat — though that observation is hardly reassuring. If he could, he would slow the pace of AI development “just so that we have more time to understand what’s going on”.
The current AI arms race, he explains, is a classic “prisoner’s dilemma” — one of the foundational parables of game theory. In the standard scenario, two prisoners must separately decide whether to confess to a joint crime. The smartest option, counterintuitively, is to confess, because neither knows what the other will do. By the same logic, AI companies are locked in competition to get ahead, spending ever more on resources and energy-hungry datacentres with no net increase in benefit for humanity. “We’ve got a small number of very, very wealthy companies that are busy pursuing AI, while at the same time saying that they are afraid that something’s going to go horribly wrong with it. So why are they busy pursuing it? Because they think if we back down and we don’t pursue it, somebody else will.”
Asked whether he was ever tempted to join Silicon Valley himself, Wooldridge says there were a few points when it could have happened, “but I’m 60 this year and it’s a young person’s game right now”. He gets many parents asking what their children should study at university. “The answer is: ‘Let them study something that they’re really passionate about.’ I think that’s the most important thing by far.”
