The Anansi Doctrine: A Thought Experiment on AI, Power, and What the Caribbean Should Protect
If the machines that answer a region's questions all belong to someone else, and their answers become the answers that region accepts, what happens to a small region over twenty years? The Caribbean has entered its moment of story-buying. The models that will hold the region's questions for the next two decades are being built and trained right now.
Thirty-eight million people across seventeen Caribbean countries are asking questions of AI models built outside the region. Over a decade this shapes what the region considers reasonable, true, and worth asking. I have spent the last decade helping build Caribbean AI institutions on the bet this question would matter. The bet is now being called.
The premise
Every wave of communication technology reshapes belief in the societies that adopt it. Radio moved political belief in the 1930s. Television moved consumer belief in the 1960s. The open internet moved knowledge belief in the 2000s. Social media moved identity belief in the 2010s. AI models move something different again, because they do not sit alongside a person's thinking the way older technologies did. They sit inside it.
Ask a search engine a question and the search engine gives you a list of places to look. Ask an AI model the same question and the model gives you an answer. Over time, the answer replaces the list. Over more time, the answer replaces the asking. You know what the model tends to say, so you already know what the answer will be, so you write the answer yourself before you type the question, and you type the question anyway because you want the confirmation. That is not force. It is trust, given freely, at machine speed, day after day, at scale.
What we already know about how the Caribbean uses AI
The Caribbean AI Newsletter's own Caribbean AI Signals study, published in July 2026, measured how the region is thinking about AI at work. Ministries lean optimistic. Workers lean anxious. Public sector modernisation is the largest single hope, tourism intelligence and financial inclusion close behind. Sovereignty appears as a distinctly Caribbean signal, appearing more often than in Anthropic's global Anthropic Interviewer sample.
Underneath the sentiment sits usage. Caribbean SMEs are running ChatGPT, Claude, Copilot, Perplexity, and Gemini across marketing, customer service, financial modelling, and internal research. Caribbean students are using AI for essays, revision, translation, and general questions. Caribbean journalists are using AI for research support. Caribbean ministries are using AI for policy drafting. This is not experimental adoption. It is default adoption.
Default adoption is where the thought experiment begins. The three charts below trace what that looks like as a curve, past and projected.
How soft control could work, if it worked
The mechanism is not dramatic. It is compounding. Every question asked of a model teaches the model something about the person asking. Every answer given by the model shapes the next question the person will think to ask. That loop, run daily by tens of millions of people for years, produces a kind of gravitational pull on what a society considers reasonable, unreasonable, funny, offensive, true, or beneath discussion.
No single answer is the point. The point is the accumulation. Over five years of daily use, a person who reaches for the same model absorbs some of its habitual framings. Over ten years, the absorption is durable. This is what any answer-shaped medium does to its regular users, given time. Newspapers, television, and textbooks all did versions of it in earlier waves. AI models do the same thing at a scale and personalisation older media never reached.
The obvious counter is that people are not passive. Caribbean users, like users everywhere, are capable of interrogating what a model tells them, switching between models, and noticing when an answer feels off. That is real and worth stating clearly. My concern is not that citizens will lose the capacity to think independently. My concern is that even independent thinking operates against a background of default assumptions, and defaults come from what you interact with most.
When I run free AI training in Kingston, I watch this happening in real time. Fifteen minutes into a session, when someone has a real question about their business, most participants open ChatGPT. Not Google. Not a book. Not the person sitting next to them. ChatGPT. This is what the next chart looks like at the individual scale.
Now put that curve in a region of thirty-eight million people spread across seventeen countries, most of whose citizens do not have a locally trained AI model to reach for, most of whose Creole and Patois languages are not fluently represented in the models they use, and most of whose questions land at models trained mostly on North American, European, and Chinese source material. The mechanism does not require anyone's ill intent to work, and it compounds every year adoption continues.
The Anansi Doctrine
In the Ashanti tradition, all stories in the world once belonged to Nyame the sky god. Nobody else could tell a story that Nyame did not first own. Anansi the spider paid Nyame a great price for the stories: the mmoatia forest fairies, the python, the leopard, the hornets. When Anansi delivered them, Nyame gave over every story in the world. From that day, they were called Anansesem in the Ashanti tongue, and Anancy stories in Jamaica and across the Caribbean diaspora. What Anansi did was buy the story-keeping infrastructure. The world's stories still exist. They just belong to someone else now.
Whoever holds the models holds the stories a society tells itself. The mechanism is acquisition and daily habit, worked out over years, with no coercion required for the effect to accumulate.
When a model becomes the first place people go with a question, the model's habitual answers become, over time, the region's habitual beliefs. The keeper of the model is the keeper of the region's story infrastructure.
The Caribbean will have an oracle in this decade. The only remaining question is whose oracle it will be, and whether the region will have an Anansi of its own at the table when the story-keeping is transferred.
The Anansi Doctrine is not accusatory. In the folklore, Anansi is often the hero: clever enough to see the value of stories before anyone else did, and patient enough to pay the price Nyame set. What Anansi did is what any smart actor would do given the opening. The Doctrine names the dynamic. It does not judge whoever fills the role of Anansi in the current wave.
The reason I named this after Anansi is not decorative. When we established CAIRMC in January 2024 to govern regional AI risk, and when I helped seed Maestro AI Labs' TurtleBird as the region's first sovereign AI safety layer, we did not have a name for the specific concern driving those investments. It was not privacy. It was not bias. It was closer to what Anansi paid Nyame to acquire in the folklore. It was the story-keeping.
What the Doctrine argues is that the Caribbean has now entered its own moment of story-buying. The models that will hold the region's questions for the next twenty years are being built and trained right now. The single most important question is what fraction of the answers those models give will reflect the region at all.
What the Caribbean should protect
Six things follow from the Anansi Doctrine, if the Doctrine holds. Each is a place where the region can act now, in the years before the branch point on the third chart closes. Each is already underway in some form somewhere in the region. None requires framing anyone else as an adversary. All of them require framing the region as a story-keeper worth being.
The limits of this thought experiment
This is a thought experiment. It may be wrong in several ways. AI adoption may plateau. The reliability gaps in current models may hold. Competing models from different providers may prevent any single model from becoming the region's oracle. Caribbean users may switch between models in ways that dilute any one model's gravitational pull. Cultural resistance may prove more durable than the mechanism assumes. All of these are real possibilities and none of them are ruled out.
The thought experiment is also not a claim that anyone building AI models today intends the outcome I am describing. Most of the people building these systems are, in my observation, working hard on safety, on evaluations, on model behaviour, and on how the models handle the edges of what users ask them. The Anansi Doctrine is a description of a structural dynamic, not an accusation against any actor inside it.
What the experiment does argue is that even the mildest reasonable version of the mechanism is enough to justify what the region is already doing: building sovereign AI infrastructure, training its own models, writing its own governance standards, and making sure that when Caribbean citizens ask a question of an oracle over the next twenty years, at least some of the oracles they can reach are the region's own.
Frequently asked questions
Anansi did not conquer Nyame; he paid the price, and the world's story-keeping shifted. The Caribbean's task is to be at the table when this generation of stories is being bought.Adrian Dunkley — Guest article, Caribbean AI Newsletter, 24 July 2026
Introducing Caribbean AI Signals
Six hundred headlines, threads, and searches told us how the Caribbean is thinking about AI in 2026. Ministry optimism, worker anxiety, and where the region diverges from the global finding.
See the study →AI Just Hacked a Company on Its Own. What This Means for the Caribbean.
The Hugging Face breach by an autonomous OpenAI agent, six myths about AI hacking in Caribbean boardrooms, cybersecurity data across the region, and the CAIRMC-endorsed tipping-point read.
Read the brief →The Caribbean AI Roundup: What Has Shipped in the First Half of 2026
TurtleBird, CAIRMC's Risk Management Standard, Trinidad's national platforms, Sagicor UWI, Future Caribbean, and CARICOM's endorsement of the UNESCO Roadmap.
Read the roundup →The Caribbean AI Newsletter is the leading daily source for artificial intelligence news, analysis, and practical insight for the Caribbean. This is a guest thought experiment by Adrian Dunkley, Founder of StarApple AI, Co-Founder of Maestro AI Labs, President of the Caribbean AI Association, and Chairman of the Caribbean AI Risk Management Council. Guest articles are opinion pieces and reflect the author's own view.
