Everyone knows increasingly powerful AI could create enormous benefits. But if everyone understands the risks, why is everyone still racing to build it?
We know nuclear weapons are extremely dangerous.
A nuclear bomb can destroy a city within seconds. Radiation can affect people and the environment for generations.
And yet, countries around the world have spent decades developing, maintaining and expanding nuclear capabilities.
Why?
Because there is a powerful argument behind the nuclear race:
“If we don’t build it, someone else will.”
And that creates a strange situation.
A country may believe nuclear weapons are dangerous. It may even wish the world had fewer nuclear weapons.
But if another country has them, giving up its own capability can feel impossible.
Security becomes competition.
Competition becomes an arms race.
And eventually, everyone is running because nobody wants to be the person who stops first.
What if AI is heading toward something similar?
The AI race has already started
Artificial intelligence is not a weapon in the same sense as a nuclear bomb.
That comparison has limits.
AI can also be enormously beneficial.
It could help researchers discover new medicines, accelerate scientific research, improve education, assist doctors, solve engineering problems and help humanity deal with challenges that are currently beyond our capabilities.
That’s the exciting side of AI.
But there is another side we don’t talk about enough.
The race to build increasingly capable AI is happening at extraordinary speed.
Companies want better models.
Countries want technological leadership.
Businesses want productivity gains.
Investors want the next breakthrough.
And nobody wants to be the company or country that falls behind.
The objective is no longer simply:
“Can computers answer questions?”
We’re moving toward systems that can reason, write code, analyse information, use tools, conduct research, create content and complete increasingly complex tasks.
And beyond today’s systems is the idea of AGI — artificial general intelligence: AI capable of performing a broad range of intellectual tasks at a level comparable to, or potentially beyond, humans.
We don’t know exactly when or whether AGI will arrive.
But the race toward increasingly capable AI is already real.
And that’s where things become interesting.
The promise of AI is enormous
Let’s be fair.
AI could become one of the greatest technologies humanity has ever created.
Think about medicine.
There are diseases we still don’t fully understand.
There are treatments that take years to develop.
There are enormous amounts of medical and biological data that humans cannot realistically analyse manually.
What if increasingly capable AI can identify patterns that researchers miss?
What if it helps discover a treatment that would otherwise take another 20 years?
The same applies to science.
Climate modelling.
Materials science.
Space exploration.
Energy.
Engineering.
Drug discovery.
Education.
AI could become an intellectual amplifier for humanity.
Instead of one researcher being able to explore 100 possibilities, perhaps AI allows that researcher to explore 100,000.
That sounds incredible.
And it is.
But technology doesn’t only change what humans can do.
Sometimes it changes what humans are needed to do.
And that’s where the uncomfortable questions begin.
What happens when AI stops being an assistant?
Today, we mostly think about AI as an assistant.
We ask it to write an email.
Create an image.
Summarise a document.
Write code.
Analyse data.
Generate a marketing strategy.
Create a presentation.
That’s already impressive.
But imagine AI becoming capable of independently handling an entire workflow.
Instead of:
Human → AI → Human → AI → Human
It becomes:
Goal → AI → Result
You tell an AI:
“Build me a marketing campaign for this business.”
And instead of giving you a few ideas, it researches the market, identifies customers, analyses competitors, creates the website copy, produces advertisements, generates images and videos, launches campaigns, analyses performance and continuously improves the strategy.
The human doesn’t disappear.
But the amount of human labour required could shrink dramatically.
Now multiply that across thousands of industries.
That’s the bigger question.
AI won’t take your job. Someone using AI will.
We’ve heard this sentence many times.
And there is truth in it.
Technology has always changed jobs.
The internet changed marketing.
Automation changed manufacturing.
Software changed accounting.
Search engines changed research.
Smartphones changed communication.
But AI could be different because it is not only automating physical tasks.
It is increasingly automating cognitive tasks.
- Writing
- Programming
- Research
- Analysis
- Design
- Customer service
- Marketing
- Translation
- Administration
And potentially many tasks that were historically considered “knowledge work.”
So here’s the question I keep coming back to:
What happens if AI doesn’t just assist the worker, but eventually becomes capable of performing most of the worker’s job?
And then another technology enters the equation.
Robotics.
AI can provide intelligence.
Robotics can provide physical action.
Put the two together and the potential becomes much bigger.
An AI system doesn’t need to physically build a house today.
But what happens when AI systems can control increasingly capable robots?
- Manufacturing
- Warehousing
- Transportation
- Construction
- Agriculture
- Delivery
- Maintenance
The boundary between “software automation” and “physical automation” starts disappearing.
Then what are we educating humans for?
This might be one of the biggest questions of the AI era.
For generations, we’ve been told:
Get a good education.
Learn valuable skills.
Go to university.
Build expertise.
Get a good job.
Build a career.
But what happens if the skills we spend 15 or 20 years learning become increasingly automated?
I’m not saying education becomes useless.
Quite the opposite.
Education may become more important than ever.
But perhaps the purpose of education changes.
Maybe memorising information becomes less valuable.
Maybe asking the right questions becomes more valuable.
Maybe creativity, judgement, leadership, critical thinking and human relationships become more important.
Or maybe we haven’t even figured out what the most valuable human skills will be yet.
And that’s the problem.
We’re preparing today’s children for a future that we don’t fully understand.
What happens to the value of human labour?
This is where the discussion becomes much bigger than jobs.
Let’s imagine a future where AI and robotics can perform a huge percentage of economically useful work.
Machines can produce goods.
AI can produce software.
AI can create content.
Robots can move products.
Automated systems can operate businesses.
AI can analyse markets.
Machines can build things.
If machines can produce more and more of the world’s output, then we have to ask:
What happens to the economic value of human labour?
Today, money is closely connected to work.
You work.
You get paid.
You use that money to buy things.
But if technology dramatically reduces the amount of human labour needed to produce those things, the relationship between work and income could become very different.
And then we face an even bigger question:
What happens to the meaning of money in a highly automated economy?
I’m not claiming I have the answer.
I don’t.
I don’t think anyone does.
But I think we should be discussing the question before we reach that situation.
The dangerous part may not be AI itself
Here’s where I think the nuclear analogy becomes interesting.
The biggest problem may not be that one company wants to build powerful AI.
It may not even be that one country wants to become the global leader in AI.
The bigger problem could be competition itself.
Imagine one country slows down its AI development because it is concerned about safety.
Another country continues.
The second country develops more powerful systems.
It gains an economic and strategic advantage.
What happens next?
The first country may decide:
“We can’t afford to fall behind.”
So it accelerates.
Then another country sees this acceleration.
It accelerates too.
Companies behave in a similar way.
If one company believes its competitor is six months away from a major breakthrough, it doesn’t necessarily want to slow down and spend another five years debating the risks.
It wants to win.
And suddenly the question isn’t:
“Should we build this?”
It becomes:
“How quickly can we build it?”
That is how races happen.
We’re still having fun with AI
And honestly, most of us aren’t thinking about any of this.
We’re having fun.
- Turning photos into 80s-style images.
- Creating Ghibli-style artwork.
- Writing LinkedIn posts.
- Writing emails.
- Generating code.
- Creating videos.
- Analysing spreadsheets.
And there’s nothing wrong with that.
Technology should be fun.
But it’s easy to underestimate a technology while we’re experiencing only its early stages.
ChatGPT became publicly available in late 2022.
Think about how different AI feels today compared with those early days.
In just a few years, we’ve gone from:
“Ask AI a question.”
to:
“Give AI a task.”
That’s a major shift.
The next shift could be:
“Give AI a goal.”
And then AI figures out the steps.
That is a fundamentally different relationship between humans and machines.
What if everyone is running because nobody wants to stop?
This is probably the part that worries me most.
I don’t think every technology leader is secretly trying to destroy humanity.
I don’t think AI is automatically evil.
And I don’t think we should simply stop developing AI.
There are enormous potential benefits.
But imagine being the CEO of a company that believes a competitor is close to achieving a major AI breakthrough.
You have two choices.
Slow down because you’re concerned about the risks.
Or continue because if you stop, someone else may get there first.
Now imagine every major company making the same calculation.
And every major country doing the same thing.
Nobody has to believe the race is completely safe.
They just have to believe that losing the race would be worse.
That’s enough to keep the race going.
And that is what makes it uncomfortable.
The question isn’t “Will AI change the world?”
It already is.
The better question is:
How much of the world are we willing to change before we fully understand the consequences?
AI could help humanity solve some of its biggest problems.
It could accelerate medicine.
Improve education.
Increase productivity.
Help scientists make discoveries.
Give individuals access to capabilities that were previously available only to large organisations.
The upside is enormous.
But so is the uncertainty.
We don’t know exactly what happens when intelligence becomes cheap and abundant.
We don’t know what happens when AI can perform most digital knowledge work.
We don’t know how AI and robotics will reshape employment.
We don’t know how governments will respond.
We don’t know how wealth will be distributed.
We don’t know what education should look like.
And we certainly don’t know what happens if AI systems eventually become significantly more capable than humans across a wide range of tasks.
That’s why the conversation matters.
Not because AI is evil.
Not because technology should be stopped.
But because the decisions we’re making today could shape the next several decades.
Maybe the real AI risk is the race itself
The nuclear age taught humanity something uncomfortable.
Sometimes the biggest danger isn’t simply the technology.
It’s the competition around the technology.
Everyone can understand the risks.
Everyone can acknowledge the possibility of disaster.
And yet everyone can still decide:
“We have to keep going because someone else might get there first.”
Maybe AI will never become the existential threat some people fear.
Maybe it will become humanity’s greatest invention.
Maybe it will help us solve problems that currently seem impossible.
I hope that’s what happens.
But I don’t think hope is enough.
Because if we’re building increasingly powerful systems, we need to think about more than just:
“What can we build?”
We also need to ask:
“Should we build it?”
“How should we control it?”
“Who gets to control it?”
And perhaps the most important question:
Are we building the most powerful tool humanity has ever created because it will make humanity better — or because we’re afraid someone else will build it first?
That’s the question I think we should be talking about now.
What do you think?
