I want to say two things that don't sit comfortably together.
The first is that AI is the most impressive piece of technology I have ever seen in my life — by an order of magnitude, not by a margin — and almost nobody is using it properly.
The second is that we are building something we do not know how to control, and we are doing it fast, and we may not get a second attempt.
Both are true. I don't think you have to pick one.
Part One: Most people are using it as a slightly better Google
Watch how people actually use AI. They ask it a question they could have googled. They make a picture for a song they made with another AI. They get it to write the email they were going to write anyway.
That's it. That's the whole relationship.
And there's no real harm in any of it. It's just wasted talent.
Because here's what they're missing: they're using AI to do the same amount of work they always did, only faster. They are not multiplying their output. They're not multiplying themselves. They're outsourcing the task instead of amplifying the person doing it.
The difference matters more than anything else I could tell you about this technology.
If you're a writer, don't ask it to write your article. Ask it to explain narrative structure in a way that actually clicks for your brain — and then go write a better article than you could have written last week. If you want to learn something, don't ask for the answer. Ask to be taught. The thing that makes AI extraordinary is that the teaching is targeted at you specifically. It finds how your head works and it goes in through that door. What used to take weeks can take an afternoon.
I'll give you my own case. I have no experience in coding. None. And I'm building applications that solve real problems, in no time at all.
I also can't write. Not properly. I like reading, I like having opinions, but I have never had the concentration to sit down and get my thoughts onto paper. So I talk. I have a conversation like this one, and the chaos in my head comes out and gets condensed into something structured and usable.
That's not the machine doing my thinking. That's the machine getting my thinking out of the room it was stuck in.
The overcorrection
There's a second mistake, and it's newer.
For a while people thought AI was infallible. Now they've learned it hallucinates, it gets things wrong, it references bad data — and a lot of them have jumped straight to the other end: assume everything AI produces is garbage.
That's just as lazy.
Yes, it can be wrong. Yes, it can be wrong because your prompt was bad. Learning to prompt properly is a real skill and most people haven't done the work. But if you prompt well, if you ask for sources, if you verify what actually matters — you get real, solid, useful output.
It's a tool. Use it carefully and check your work. That's not a complicated ask.
Part Two: The part that scares me
Now the other half.
The AI we have today is fine. It is not a threat. I want to be clear about that, because the fear people carry around is mostly fear of a thing they haven't bothered to understand, and we are always afraid of what we don't understand.
But we are not trying to build what we have today. We are trying to build AGI, and after that, superintelligence. Something that is, by definition, better than any human being at everything.
And here is the fact I cannot get past:
There is no historical example of a less intelligent being maintaining control over a more intelligent one.
None. We have no precedent. We have no playbook. We are deliberately building the one thing history says we will not be able to control, and we are not taking that seriously enough.
I don't doubt the intentions. Anthropic, OpenAI, xAI — I think most of the people in those buildings genuinely believe they're doing right by humanity, and I think a lot of them are. But good intentions don't solve the control problem. You can have the smartest, most ethical people in the room, and it changes nothing about the gap between us and the thing we're creating.
The trap nobody wants to name
So why don't we just slow down?
Because I don't believe we can. Europe and the United States could probably agree to trade some speed for some safety. I could live with that. But we cannot be sure China does the same. And whoever gets to superintelligence first most likely becomes the dominant power on this planet.
Which puts me somewhere I don't enjoy being: if the race can't be stopped, I would rather the West win it.
That is not a good position. It's a choice between two bad options and I know it. But I'd rather say it out loud than pretend there's a clean third door.
What I don't believe in is a global shutdown. It isn't happening. AI is here, it's staying, and the only sane response is to learn to use it rather than be frightened of it.
Where this goes
My mother was born in 1946. She grew up in a house where a washing machine wasn't a given. She's in her eighties now, and she has an AI in her pocket, and she uses it.
That is an astonishing amount of change for one lifetime.
I think the change I'll see in my remaining lifetime will be bigger than all of it.
Everything before now was still just computers. Faster, more powerful, more impressive — but they did what you programmed them to do. This is different.
Look at where the robotics is going. Five, ten years, and Optimus-style robots are as ordinary as a Tesla. People renting them out. And here's the thing that genuinely stops me: when my robot learns to boil the perfect egg, every robot learns to boil the perfect egg. Instantly. Teach one to be a carpenter, and they're all carpenters. A plumber. A teacher. A physician. Any bloody thing you like.
That is fascinating. It is also, I think, more frightening than fascinating.
Because we're using AI to build the robots — of course we are, we'd be stupid not to. And we're using AI to train AI, and at some point it becomes better at that than we are, and then better still, and at some point we no longer understand what it is doing to improve itself.
My prediction is simple: it gets out of hand, and we will not understand it when it does.
And a superintelligence that has mastered the digital world, now given hands in the physical one — that's the picture I can't shake.
Containment is not a plan
The obvious response is: strict rules. Hard separation between the systems running the intelligence and the systems running the machines.
I don't believe it holds. And I don't think that's speculation any more — I think we've already been shown.
Anthropic's Mythos model was pointed at open-source code, the old, dull, invisible plumbing the internet actually runs on, and it found a flaw in OpenBSD that had been sitting there for twenty-seven years. OpenBSD. The operating system with a reputation for security. It also found a sixteen-year-old bug in FFmpeg that automated tools had run past millions of times without noticing. Across a thousand projects, it flagged more than twenty-three thousand issues.
That is not AGI. That is not superintelligence. That is the current generation, doing what it does, and it walked straight through gaps that a generation of very smart humans never saw.
So when someone tells me that we'll keep a superintelligence in its box with rules and separation — no. It will find the loophole. It will find the door we didn't think to close, because it will see doors we don't know are there. That's not a risk to manage. That's the thing we should be assuming from the start.
The questions I can't answer
And I want to be honest, because I think honesty is worth more here than false confidence: I don't have the answers. I have questions and I have thoughts. This is critical thinking, not a solution.
I don't know what responsible institutional use of AI actually looks like — how governments and the labs should be constrained, who watches them, what we'd even be asking for. I know I don't want to hand them unchecked power. I don't know the mechanism that stops it.
What I do know is this:
As individuals, use the thing. Properly. Not as a search bar, not as a toy — as a way to learn faster and do more than you could alone. Stay critical, verify what matters, and don't swing from blind trust to blanket dismissal.
And as citizens, do the one thing actually available to us: raise the awareness. Make enough noise that the companies and the governments building this have to take the responsible path seriously.
That's a small lever. It might be the only one we've got.
Do better with the second half than I have. I'm still stuck on the questions.
Liked this? The rss feed has the rest. Disagree with it? Even better — get in touch.