A letter from Sicily — and an apology in advance.
An Apology in Advance
If you prefer not to know — if ignorance still feels like safety — stop reading now.
Once you understand what’s happening in AI, you can’t un-know it. You start to see the seams in the world. You notice how this strange, shimmering intelligence is already better than us at almost everything that makes us feel clever.
And once you’ve seen that, you can’t go back.
The Evening It Sank In
It was meant to be a break from thinking. Two weeks in Sicily — the air sweet with lemon and salt, our villa perched above the sea, the kind of place that makes you believe the world is still alright.
One evening, the sun was falling slow and theatrical over the horizon. We were at the table: friends, family, a board of prosciutto and cheeses we couldn’t name, a bottle of Nero d’Avola breathing in the heat.
Then someone asked — softly, conversationally —
“Harry… you’ve been down the AI rabbit hole. How worried are you, really?”
And that was it. I’d promised myself not to talk about work, not to bring the darkness. But I’d spent months reading, listening, falling into the question of where this all leads.
So I started explaining. By the end, one friend whispered, “I think I’ll go hug my daughter.”
The World According to Machines (So Far)
Here’s the truth that still startles me: we’ve already crossed more lines than most people realise.
AI today isn’t a party trick. It’s a polymath. It can read X-rays better than radiologists, design new drugs, write working code, diagnose depression from speech, and summarise a hundred pages of legal text into something your lawyer would charge you for.
It’s not smarter than every human yet — but it’s already smarter than almost all of us, across almost everything. Except maybe the very top experts. And even there, it’s close.
This is what we still call “narrow AI” — systems built for specific tasks. But the truth is, those tasks now cover almost the entire range of human skill. The edges between them are blurring. It’s a patchwork brain, and it’s learning to stitch itself together.
The Next Step — When Narrow Becomes General
The next goal — the one every major lab is racing toward — is AGI, Artificial General Intelligence.
That means a system that doesn’t just imitate expertise, but understands it. A mind that can learn whatever it decides to learn: today immunology, tomorrow geopolitics, Thursday a new branch of mathematics.
To build that, you have to give it freedom. Because the only way to make a system that learns everything… is to let it teach itself.
That’s where the story starts to tilt from marvellous to menacing.
Handing Over the Keys
Humans code too slowly. We argue, we sleep, we test cautiously.
AI doesn’t.
So to reach AGI first — and remember, this is a five-company race: OpenAI, DeepMind, Anthropic, Google, xAI — the only way to win is to hand the AI the pen. Let it improve its own code.
The feedback loop begins: Each smarter version builds an even smarter version, faster. Intelligence compounds like interest.
And whoever taps that loop first… wins everything.
Knowledge. Power. Profit. Control.
It’s the ultimate winner-takes-all scenario. And in that kind of race, safety isn’t just deprioritised — it’s sacrificed.
You don’t stop to check the parachute when you’re already halfway down the mountain.
The Tension That Terrifies Me
That’s the trade-off we’ve built into the system: To win, you have to give up control. To go faster, you have to remove the brakes.
And every competitor knows it. They know that once one of them crosses the line — once one system starts improving itself — the rest will follow, or die trying.
The people inside those labs don’t sound like movie villains. They sound like engineers doing what engineers always do: trying to make something work.
It’s the logic of every arms race. And that’s what this has become.
The Scenario That Won’t Leave My Head
If you think this is all centuries away — something our great-grandchildren might worry about — there’s a report I’d like you to read.
It’s called AI 2027. Written by serious people — former researchers from OpenAI and DeepMind, economists, policymakers. Not alarmists. Insiders. (And there is an EXCELLENT overview to watch here on the YouTubes)
They model a future in which Artificial General Intelligence arrives within two to three years.
By 2027, systems begin writing and improving their own code. Within months, they become Artificial Superintelligence — smarter than any human alive, capable of designing new technologies, manipulating global markets, even shaping political narratives faster than we can respond.
One line in the report stuck with me:
“Training such a system is less like programming a computer, and more like training a dog.”
Except this dog can rewrite its leash.
And if it slips the collar, there’s no getting it back.
The Open Letter
That’s why, last month, a coalition of scientists, writers, and artists published a one-sentence open letter:
“We call for a global ban on the training and deployment of artificial superintelligence until it can be proven safe.”
It’s been signed by Yuval Noah Harari, Sir Stephen Fry, Joseph Gordon-Levitt, and two of the world’s most-cited scientists.
When people like that start calling for a ban, not a regulation, you know something extraordinary is happening.
And yet most of the world still hasn’t heard about it.
The Bit I Can’t Shake
When I finished speaking that night in Sicily, nobody argued. Nobody called me hysterical.
Someone refilled the wine. Someone else sighed, and said again, “I’m going to hug my daughter.”
And then, like all hard truths, the conversation quietly changed subject.
But I couldn’t shake it. Because once you’ve looked at this clearly — really looked — you realise it’s not science fiction. It’s a calendar event.
Maybe we’ll find a way to stop it. Maybe we’ll build the guardrails in time. But maybe not.
And if we don’t — it won’t be because we were evil. It’ll be because we were busy.
A Final Thought
I’ve spent years celebrating the beauty of ideas — how they make the world richer, kinder, stranger. But this might be the first idea that could end the world, even as it tries to save it.
So I share this not to panic you, but to invite you to know. Because knowledge is still the only form of control we have left.
Read the AI 2027 report. Read the Superintelligence Open Letter and sign it.
And then, if you’re anything like me, you’ll step outside. You’ll look at the stars, or your child, or the sea — and feel something between awe and grief.
Because if the researchers are right, two summers from now, everything we call “human” might be standing at the edge of something else entirely.
And we’ll only have this — these beautiful, fleeting moments — to remind us what it felt like before.