For most of the internet age, one of technology’s great promises was access.
Give more people access to information, and opportunity should spread with it.
The smartphone pushed that idea much further. Suddenly billions of people carried maps, encyclopaedias, cameras and an advanced form of global communication in their pockets.
AI appears to take the next step.
Now we are not simply being given access to information, although on the surface it may appear that way. We are being given something far greater: something that can help us interpret, organise, question and act on information.
That sounds enormously democratic.
And it may well be.
But I keep coming back to a simple thought:
Two people can have the same AI and achieve radically different outcomes.
They can use the same model, on the same laptop, with access to broadly the same intelligence.
One might use it to build a business.
The other might use it to write a slightly better email.
Equal tools do not mean equal outcomes.
That difference interests me because access to intelligence and the ability to turn intelligence into something useful are not the same thing.
There is a gap between the two.
I think we are going to hear much more about it.
Access was only the first problem
We already know what a digital divide looks like.
At first it meant access to computers, then broadband, then smartphones.
As those technologies spread, the conversation moved towards digital literacy. It was no longer enough to own the technology. You needed some idea of how to use it.
AI simplifies and also complicates that distinction.
One of its most remarkable qualities is that it lowers technical barriers that existed before.
You don’t necessarily need to know how to code to make software, or know every formula to analyse a spreadsheet. You can create something visual without spending years learning professional design tools.
You can explain what you are trying to do and let the machine help with the difficult parts.
That should make people more capable.
But it does not follow that it will make people equally capable.
Same machine, different person
Imagine two people using AI to assess a business idea.
One asks it to write a business plan and receives a respectable-looking document.
The other explains the location, rent, available capital, likely customers, local competition, margins and staffing costs.
Then they keep going.
They question what they are being told, test assumptions and bring their own experience into the conversation.
The AI may have done a great deal of work for both.
But the quality of the outcome is not determined by the machine alone.
The person matters, and while that sounds obvious when written down, I’m not sure we have fully absorbed what it means.
Capability inequality
The phrase capability inequality is not new.
It has an academic history, particularly around the Capability Approach associated with Amartya Sen and others. Broadly, that tradition asks us not only to look at what resources people possess, but what they are actually able to do with them.
That feels especially relevant now.
Researchers are beginning to see something similar with AI.
A 2026 study by Lihi Idan and Bharat Anand, Generative AI and the Productivity Divide: Human-AI Complementarities in Education, described AI Interaction Competence: the ability to get useful results from an AI system, judge those results and verify them.
The interesting part was not simply that AI improved performance.
It was that people did not benefit equally.
Some extracted far more value from the same technology than others.
If access to AI becomes almost universal, access itself may stop explaining very much.
What matters is what happens next.
The value may move elsewhere
We tend to talk about AI as though intelligence itself is becoming abundant.
There is some truth in that.
Things that once required specialist knowledge, expensive advice or hours of research can now sometimes be done in minutes.
But when something becomes abundant, value often shifts elsewhere.
If answers become cheap, knowing which questions are worth asking becomes more valuable.
If information becomes easy to obtain, judgement matters more.
If almost anyone can produce something competent, perhaps taste does too.
The qualities that may matter most in an AI-rich world may turn out not to be especially technical at all.
Curiosity. Judgement. Scepticism.
And none of these are distributed equally.
Curiosity suddenly has much more power
Curiosity is a good example.
Somebody asks one question and moves on.
Someone else asks another.
Then another.
What am I missing?
What would somebody who disagrees say?
Curious people have always had an advantage. The difference is that curiosity used to be expensive.
Following ten lines of enquiry might mean finding ten books, speaking to several people or spending days researching.
Now the next question costs almost nothing.
That changes the economics of curiosity.
A relatively small difference between two people can be amplified very quickly when one of them naturally keeps digging.
The great equaliser may also widen the gap
This is the strange part.
I do think AI can be enormously equalising.
A child can receive an explanation tailored to them. Someone without technical training can build something they could never have built before, or a small company can perform work that once required a much larger organisation.
Those are real gains.
But there is no rule saying that raising everyone’s capability must reduce the distance between people.
AI may raise the capability floor, but it may raise the ceiling even faster.
Someone who previously could not do a task may become competent.
Someone already skilled, curious and experienced may become capable of something far beyond what one person could previously achieve.
Both improve.
The gap can still grow.
Education becomes more important, not less
There is an obvious implication for education.
If a machine can increasingly provide facts, explanations and procedures, then knowing facts cannot be the entire point.
That doesn’t make knowledge worthless, because you still need enough understanding to know when the machine is wrong, enough context to judge an answer and enough grasp of a subject to ask useful questions in the first place.
Two students may therefore have access to exactly the same AI but use it very differently.
I suspect we need to think harder about what education is actually supposed to develop.
Not just recall, but reasoning, judgement and the ability to challenge a convincing answer.
Experience still counts
AI also exposes the difference between knowing about something and having lived through it.
A model can explain negotiation, but that’s not the same as having negotiated hundreds of times.
It can describe leadership, but again that’s not the same as having watched a team fall apart because you made the wrong decision.
Experience creates patterns against which the machine’s answer can be tested.
An experienced person can look at something perfectly plausible and think:
That does not feel right.
Sometimes that instinct will matter more than knowing how to prompt cleverly.
An inequality that is difficult to see
Capability inequality may also be harder to recognise than earlier forms of inequality.
Two employees can have the same job title, laptop, AI account and training.
On paper, their access is identical.
Yet over time one may become dramatically more productive because they learn how to combine the technology with what they already know.
That matters for organisations.
Buying AI for everybody does not mean everybody receives the same benefit from it.
And telling people to “use AI” may eventually sound as crude as telling someone to “use the internet.”
What comes after access?
There seems to be a progression.
First we asked:
Who has access to technology?
Then:
Who knows how to use it?
AI may introduce a more difficult question:
Who can turn access to intelligence into actual capability?
We do not yet know how large that divide will become.
Training may narrow it, as may AI itself becoming better at helping inexperienced users.
But I doubt the difference disappears entirely.
Because the machine is only one half of the relationship.
The other half is us.
The machine may not be the advantage
For the moment, having access to a particularly capable AI system can still feel like an advantage in itself.
That may not last.
The models will improve. Competitors will catch up. Access will spread.
Eventually AI may feel less like a special product and more like infrastructure.
Something everyone simply has.
At that point, perhaps we will stop asking:
What AI are you using?
And start asking something more revealing:
What are you able to do with it?
Because the great promise of AI is that extraordinary intelligence becomes available to almost everyone.
The uncomfortable possibility is that we give millions of people access to the same intelligence — and discover just how differently they are able to use it.