Julian Darley argues in interview with The Mint that artificial intelligence can be a power for good enabling us to reap the best outcomes for humanity. The real concern, he argues, is not over whether it exists, but over who gets to shape it.
Julian Darley sees artificial intelligence (AI) as a tool that could help people live better, govern better and navigate complexity more intelligently. But he acknowledges that it could also deepen monopoly, surveillance and ecological damage. He is, however, not interested in presenting himself as a pivotal player in artificial intelligence. “I don’t claim to be a central player like (physics Nobel laureate) Geoffrey Hinton in any way,” he says. But he says he was thinking about AI long before the current wave of investor frenzy, political panic and corporate marketing.
His route into AI was practical. “My early interest was from the point of view of automatic speech recognition,” he says, alongside optical character recognition. The attraction was obvious. Darley had always been “interested in the spoken word and the written word”, and he knew from experience how punishing transcription could be. “I used to say it takes at least eight hours for every one hour of recorded speech. And it’s probably longer than that.”
That was not a minor irritation. It was the kind of repetitive human labour that cries out for technical assistance. Yet for years, he says, the reality did not match the promise. “I was just a user of it and a very disappointed one, too, let me say. I mean, it was always hopeless. Never worked very well.”
I was just a user of it and a very disappointed one, too, let me say.
Darley’s own career has hardly been linear. He describes it, with some understatement, as “very strange”: “from science to music to film, pop videos, industrial films, medical films, Hollywood script doctor”, followed by academic study and a growing concern with the environmental crisis. That broad background matters, because his interest in AI is not narrowly technical. He approaches it as someone thinking about culture, ecology, systems and everyday life all at once.
The shift from disappointment to excitement came with the deep-learning era, but above all with the public arrival of large language models. “The moment of takeoff was the same moment as for many, many other people,” he says. ChatGPT 3.5 felt “practically magic”. From there, he quickly saw its potential in coding and software development. In his words, newer models increasingly mean that “you really need to not spend very much time with the actual lines of code and more shaping the whole thing. So you’re much more a conductor”.
That capacity, he argues, is already changing what one person can do. Darley writes software for his own projects and for practical tasks. One example involved prospect research: helping identify likely donors among very large lists of companies and executives. AI, he says, is not yet brilliant at swallowing huge datasets in one go, but if broken into manageable chunks it can save enormous amounts of effort. “If you do that 2,000 times over, that’s a very tough thing for a human to do.” For him, the point is not magic automation. It is an augmented capability.
Yet Darley is most animated not by coding itself but by what AI might do for human wellbeing. He breaks the opportunities into the personal, the public and the corporate. The first of these is more radical than it sounds.
On food, his case is blunt. “A very significant percentage of greenhouse gases are emitted by the food system sector,” he says. At the same time, “we have taken enormous amounts of land from nature. So nature has very little left.” If people are serious about biodiversity and climate, then reducing the impact of food matters. The simplest route, he says, is to eat “further down the food web, further down the food chain. That means eating more plant-based food.”
The obstacle is not abstract moral disagreement so much as daily life. People like meat. Many are overstretched. Many are not skilled cooks. Many do not know how to produce plant-based meals that are nutritious, appealing and affordable. Here Darley becomes unexpectedly enthusiastic. “This is one of the places where AI is going to be frankly your saviour,” he says. Why? Because it can draw on huge volumes of culinary knowledge, tailor recipes to taste and ability, and refine them in response to what the user actually wants.
“If you say… I’m a lousy cook. Can you simplify it a bit? It will,” he says. More than that, it can be instructed to optimise for flavour as well as health: “Please make it as tasty as possible, even whilst you’re making it nutritious.” For Darley, this is not trivial assistance. “You need a lot of skill and knowledge to prepare a really good, nutritious, interesting plant-based dish that will be so good that you’ll say, ‘oh, actually, I’ll have that over the steak.’”
Domestic robots are coming… and though they will generate fear and sensational headlines, they could be transformative for ordinary life.
He then pushes the argument a step further, into robotics. Domestic robots are coming, he says, and though they will generate fear and sensational headlines, they could be transformative for ordinary life. “Let’s say you’re not a very good cook or you’re disabled, whatever it is, or you’re just exhausted because you’ve had to work some stupid 12 hour a day… dead end job. And yet you still want to eat better.” In that context, AI-guided robotics could provide practical help. “In a way, we’d all have our own Cordon Bleu cook who would be able to do a five pound meal.”
From there the conversation moves to governance, where Darley’s claims become bigger and more contentious. His core argument is that modern society has become too complex for unaided human judgement. “We have a staggeringly complex world,” he says. “No human and more importantly, no group of humans can get close to understanding what the heck is going on.” Yet decisions still have to be made across energy, land, food, infrastructure and ecology, and too often they generate “yet another well-meaning bad decision, which has some horrible effect over here.”
AI, he argues, is uniquely suited to this level of systems complexity. “AI is the only thing which can deal simultaneously with millions and billions and trillions of intersecting, interleaving data points.” Without it, he says, “I don’t think there’s a hope in hell of coming up with some better decisions.”
Energy is one example. A distributed renewable grid is inherently more difficult to manage than a system based on a few giant centralised assets. “If you’ve got one point source, two gigawatt nuclear reactor, that’s much easier to deal with.” But a system of fluctuating solar, storage and local generation produces constant variation. “This is… the same kind of thing where AI is designed to cope with a staggering number of data points.” That, he says, creates “a gigantic opportunity” for smarter grid management.
Landscape regeneration offers another example, and one likely to resonate with readers of The Mint. Darley calls it “another perfect example” of a complex field full of competing interests, uncertain conditions and unintended consequences. Weather, soil, land use, human conflict and non-human life all interact. “Nature… doesn’t usually have a voice,” he says, half-jokingly suggesting “a badger AI or a badger agent” alongside the human actors. The serious point is that AI might help “in real time, help balance these competing uses and needs”.
One reason, he says, is that AI lacks ego. “It doesn’t have consciousness. It doesn’t care about itself.” In principle, that means it can weigh trade-offs without the career anxieties, turf wars and institutional defensiveness that so often distort human decision-making.
There are definitely legions of risk.
But Darley is no utopian. “There are plenty of risks,” he says. AI can reinforce addiction and “bad behaviour”. It can intensify surveillance and militarisation. Data centres consume huge amounts of water and power. The rush to build AI infrastructure is already distorting chip and memory markets. “There are definitely legions of risk.”
So the question becomes control. Darley returns repeatedly to the importance of running models locally rather than sending every query back to giant cloud companies. If prompts always go to Silicon Valley, “they keep control and they can monitor it and they can see what you’re asking”. Local models change that. “Much more difficult for them if you are sending it to a local model on your phone or on your computer and it never gets out of your computer.”
His conclusion is not that AI will automatically liberate people. It is that the fight is still open. “AI can be their greatest friend,” he says. “In the hands of certain ghastly actors… it might well become the greatest enemy.”
That is the crossroads. Darley sees enormous promise in AI, but none of it arrives cleanly. The technology may help people cook, think, model, cooperate and govern in a time of accelerating breakdown. But unless power is dispersed, and unless more people engage rather than simply recoil, it may also deepen the grip of already dominant firms. Darley’s view is clear enough: the benefits are real, but they will not deliver themselves.
