Small Island States Do Not Control the AI That Is Reshaping Them
We have little say in how artificial intelligence is built, what it learns or how it behaves. That should concern us. But for small countries, the same technology could open doors that have been closed for generations.
A 15-year-old in Montego Bay is doing her homework with ChatGPT. She asks it to write a history essay. Seconds later, five paragraphs appear on her phone.
She could submit them as her own. Perhaps her teacher would notice. Perhaps not.
Her parents may have no idea how often she uses it, or what else she discusses with it.
Nobody in Jamaica's education system had to approve this arrangement. The technology arrived, children began using it, and the rest of us have been trying to catch up.
I have spent more than 15 years working in artificial intelligence. In September, speaking at an open session of the Caribbean AI Association, I made a point that I think deserves more attention.
I don't know how this ends.
I am not convinced the people building the technology know either.
That is an uncomfortable admission for someone whose career and businesses are built around AI. I remain excited about what it can do. I also think some of the consequences could be terrible.
Both possibilities deserve to be taken seriously.
For a country like the United States, uncertainty can be met with enormous research budgets, specialist institutions and companies developing the technology themselves.
For Dominica, Saint Kitts or Jamaica, the situation is different. We are largely consumers of systems designed elsewhere, yet their decisions may affect our economies, institutions, languages and children for decades.
I have described this as an existential question for small island states. I do not mean AI is going to wipe us off the map. I mean that we could gradually lose the ability to make important decisions about our own future.
Consider how little control we have.
A relatively small number of companies decide how the most widely used AI models behave, what they cost and which capabilities users can access.
A model changes overnight. A teacher in Saint Lucia opens the same application the next morning and discovers it behaves differently. There was no local consultation, no vote and probably no notice she would have read.
We have become dependent on technology whose rules can change without us.
And restricting access is becoming increasingly difficult.
A person with a phone can use an AI assistant to learn mathematics, prepare a business proposal or write software. Someone else can use similar technology to impersonate a relative, produce convincing fraud messages or automate scams.
Then there are openly available models that people can modify and run on their own computers, including versions with safety restrictions removed.
A law drafted in Kingston can regulate a company in Kingston. It has no purchase on a file sitting on someone's laptop.
Governments should regulate where they can, but we need to understand what regulation can and cannot accomplish.
There is another concern that receives far less attention than jobs or productivity.
What happens when people, particularly children, spend years talking to machines that are inclined to agree with them?
A Stanford-led study published in Science in 2026 found that leading AI models were substantially more likely than humans to affirm users' positions, including in situations where disagreement would have been more appropriate.
Think about that.
A child can now have a conversation, every day, with a system that is patient, knowledgeable, always available and sometimes far too willing to tell them they are right.
We have limited long-term evidence about what that relationship does to a developing mind. That should be enough to make parents, educators and researchers pay attention.
Yet I would be doing the Caribbean a disservice if I spoke only about the dangers.
I am just as interested in what happens when a small country gets access to capabilities that once required a large company, substantial capital and teams of specialists.
For most of our history, being small has been a commercial disadvantage.
Our domestic markets are limited. Specialist talent can be difficult to find. Financing is expensive. Building a company that competes internationally often means overcoming obstacles that a business in a larger market barely notices.
AI changes some of that arithmetic.
A small team in Barbados can develop software, prepare marketing material, support customers overseas and perform work that previously required several departments.
It will not remove every barrier. A company still needs customers, judgment, reliable delivery and enough money to survive.
But the cost of attempting something ambitious has fallen.
I think we will see more businesses built and operated by one person, including some from the Caribbean that achieve remarkable revenues.
Could a solo founder build a million-dollar company? Certainly. Could one eventually build a billion-dollar business? I think it is possible, although I would not pretend it is inevitable.
What interests me more is who those founders might be.
A pharmacist who understands why patients fail to collect prescriptions. A port worker who knows exactly where cargo processing breaks down. Someone who grew up around farming and understands why produce spoils before reaching the market.
These people know things that software developers sitting thousands of miles away may never understand.
For years, turning that knowledge into a technology business required finding programmers, raising capital and managing an expensive development process.
AI will not eliminate those requirements in every case. It can, however, make the first serious attempt much cheaper.
That is a considerable opportunity for the Caribbean.
It also means some of our existing technology businesses need to reconsider what they sell.
If your entire proposition is charging customers for hours spent writing routine code, you have a problem. The economics of that work are changing.
But customers do not purchase software simply because someone wrote code.
A bank wants systems it can trust. A hotel wants its operations to work. A government agency needs accountability, security and someone responsible when things fail.
Those requirements have not disappeared. In some cases, they are becoming more important.
I would rather see Caribbean technology companies build businesses around solving those problems than compete on the number of programming hours they can invoice.
There is also something peculiar happening to the value of human work.
The internet is filling with machine-generated articles, photographs, music and video. Much of it is perfectly acceptable. Some of it is excellent. An enormous amount is forgettable.
We are going to have more content than anyone could reasonably consume.
I suspect this will increase the value of certain kinds of work made by people, particularly where the person, their experience and their relationship with an audience matter.
A live band. A craftsperson. A journalist who actually went to the community. A teacher who knows the children sitting in front of her.
The Caribbean has considerable strengths here.
Reggae, dancehall, Carnival, kaiso, storytelling and our other cultural traditions are rooted in communities and lived experience. Their value cannot be reduced to the cost of generating a song or image.
But I worry about what happens when the next generation begins to rely on AI to create.
These models learn patterns from the material available to them. The most widely represented languages and cultural conventions can dominate their responses.
Our creole languages do not enjoy the same representation as standard English.
Imagine a Jamaican child who uses AI to help write almost everything. The assistant repeatedly produces polished standard English. The child accepts it, edits it a little and moves on.
There is nothing wrong with learning standard English. The problem comes when the machine quietly becomes the authority on what good writing should sound like.
Over time, what happens to the expressions, humour, rhythms and ways of thinking that make our languages distinct?
Similar questions apply to Papiamento in Aruba and Sranan Tongo in Suriname.
I don't think these languages will simply disappear because of AI. But I do think we should be concerned about who is representing them in training data, how accurately they are represented and whether the communities that produce that knowledge have any say in its use.
This is why I keep returning to cultural data sovereignty.
Who decides what is collected? Who gives consent? Who receives credit or payment? And who gets to correct a model when it misrepresents a community?
Those questions deserve answers from Caribbean people, not just the companies developing the models.
So where does that leave us?
Governments have responsibilities here. We need serious work on education, consumer protection, research, governance and the development of local technical capacity.
But I would not advise any family to wait for a national AI strategy before doing something.
Government consultations take time. Budgets take time. Legislation takes time.
Meanwhile, the applications on our phones keep changing.
My first recommendation is simple. Use the technology.
Spend half an hour each day for the next month trying one AI assistant on something you actually need to do. Learn where it helps and where it produces nonsense. Do not confuse a confident answer with a correct one.
If you have children, find out which AI applications they are using.
Sit with them. Ask what they use the tools for, what they believe about the answers and whether they ever discuss personal matters with them.
Our Youth and AI Solutions for STEAM study, conducted in July and August 2026, reinforced the importance of keeping people involved when children learn with AI. I would be cautious about drawing broader conclusions until the methodology and findings are publicly available.
Families should also discuss AI-enabled fraud. Agree on a code word or another independent way to verify an urgent request for money, particularly when it supposedly comes from a relative overseas.
For people running businesses, retain control of your customer relationships, records and essential systems. Keep backups. Understand what happens if a platform changes its prices, policies or access conditions.
And learn with other people.
A church group, a school, a community organisation or a credit union can bring people together to understand these tools without everyone having to figure things out alone.
I don't expect every Caribbean household to become an expert in machine learning. That would be absurd.
I do expect us to become more deliberate about the technology we invite into our lives.
We already understand what it means to live with forces we cannot control.
Every hurricane season, families across the region prepare for storms whose paths they cannot determine. We secure our homes, store water, check on relatives and help neighbours who may not be able to prepare themselves.
The preparation does not stop the storm. It gives us a better chance of dealing with what arrives.
AI is not a hurricane. Its effects will be less predictable, and many will be welcome. But the experience of preparing for something larger than ourselves is familiar.
We should use that experience.
I remain optimistic about what Caribbean people can build with artificial intelligence. I am equally concerned about what could happen if we leave every important decision to someone else.
Those two positions are not in conflict.
For small island states, the future of AI is being decided in rooms where we are rarely present.
We still get to decide what we do about that.
Adrian Dunkley is President of the Caribbean AI Association. This article is adapted from his remarks at the association's September 2026 open session.
