Whose Values Get Coded? An Ethics Lens on AI Decisions in the Caribbean
Every AI system is a values document. The choice of what to optimise, what to penalise, what to surface, what to hide, whose data to learn from, whose patterns to consider normal, whose accent to recognise, whose face to enrol: each of these is an ethical choice, made by someone, applied to everyone. Most of those someones do not live in the Caribbean. Many do not know we exist. A few would struggle to find Saint Lucia, Suriname, or Saint Kitts on a map.
This article is not another general AI ethics overview. It is a Caribbean-specific lens on the ethics question, written for the boards, regulators, ministries, and citizens who will, in the next few years, have to decide what AI should and should not do in our societies. The earlier piece on ethical AI frameworks for developing nations argued for proportionate implementation of international principles. This piece goes deeper: it asks which Caribbean values should be coded into the AI systems we deploy, and what changes when we take that question seriously.
The Values Already Inside the Tools
Before we talk about Caribbean values, it is worth being clear about the values already inside the tools that are entering our economies. Three are particularly consequential.
The first is individualism. Most AI tools built for credit, insurance, hiring, and recommendation are oriented toward individual scoring. A person is a vector of features. Their relationship obligations, their household structure, their community standing, their remittance flows, their reputation in their workplace and church, are either invisible to the model or treated as noise. Caribbean economic life is built on these very obligations. Caribbean households often pool resources across multiple workers, often across multiple countries. A credit model that asks only "what is this individual's income and credit history?" misses the half of the Caribbean economic story that determines whether the loan will be repaid.
The second is the assumption of formality. Most AI tools assume the data points they ingest come from formal institutional records: a tax authority, a credit bureau, a regulator, an employer. Caribbean economic life is heavily informal. Estimates of informal employment across CARICOM run from a quarter to nearly half of total employment, depending on the jurisdiction and the definition. A tool that disregards informal income or treats it as suspicious is making a values choice that, in a Caribbean context, systematically disadvantages the people who hold the region's small businesses, agriculture, fisheries, and household services together.
The third is the universalism of taste. Recommendation engines, content moderation systems, and generative tools were trained on data that reflects, overwhelmingly, the cultural mainstream of the global north. They quietly normalise certain accents, certain naming conventions, certain religious and family structures, certain ways of speaking and writing. Caribbean creoles, Caribbean naming, Caribbean cadence, Caribbean humour, and the specific texture of Caribbean public discourse sit outside that mainstream. A content-moderation model trained on North American English speech patterns will, predictably, mis-classify Caribbean speech. A generative model asked to write in Caribbean voice will, more often than not, produce a thin parody of it.
None of this is malicious. It is a values choice made by people whose values are the ones reflected. The corrective is not to denounce the choice. The corrective is to be clear that it was made, and to decide deliberately whether we accept it for our own use.
Caribbean Values Worth Coding In
The Caribbean is not one place. There are real differences between the Anglophone CARICOM core, the Spanish-speaking Caribbean, the French and Dutch territories, the diaspora, and the indigenous communities that long pre-date all of those traditions. But there are shared regional values that recur across the surveys, the literature, and the day-to-day life of the region. Four are particularly relevant to AI ethics.
Community as a unit of moral concern. Caribbean ethics is not strictly individualist. A decision made about one person affects their household, their extended family, their church, and their workplace. AI tools that produce decisions about individuals, in a Caribbean context, should be designed with the awareness that the consequences will travel through those networks. This is not an argument for collectivism. It is an argument that the individualist default of foreign AI tooling is a poor fit for Caribbean ethical reality.
Reputation and visibility. Small island societies operate on dense networks of personal knowledge. A bank manager in Anguilla, a doctor in Bequia, a school principal in Carriacou, a customs officer in Tortola, are unusually visible. Their decisions affect people they will meet at the supermarket the next day. This visibility, well-understood in Caribbean institutional design, produces accountability in ways that anonymous large-market interactions do not. AI tools that operate behind opaque scoring systems, with no visible decision-maker and no easy avenue for personal redress, disrupt this accountability pattern. Caribbean AI deployments should preserve the visibility of the human decision-maker behind the algorithm.
Hospitality and second chances. The Caribbean tradition of welcoming the stranger, accommodating the returning prodigal, and forgiving the recoverable mistake is older than any of the modern Caribbean states. Many AI systems are unforgiving by design: a missed payment, a flagged transaction, a low engagement score, becomes a permanent feature in the data and follows the person across contexts and years. Caribbean ethics has a longstanding presumption against permanent moral records for recoverable mistakes. AI systems deployed in Caribbean contexts should be capable of expiring negative signals on a humane timeline.
The dignity of the small. Caribbean public ethics, at its best, refuses the equation of importance with size. A small farmer in Grenada, a small shop owner in Roseau, a small fishing co-operative in Antigua, are not negligible because they are small. AI tools that are designed for scale, and that quietly under-serve or mis-serve small users because the volume is not commercially interesting, sit poorly with this Caribbean value. A Caribbean AI ethics that takes the dignity of the small seriously will reject deployments that work well for the average user but fail predictably for the smallest 5%.
Five Decisions Every Caribbean Board Should Make Explicitly
Most AI ethics failures in regulated Caribbean institutions are decisions that were never consciously made. The vendor's default became the institution's policy, by accident. Five decisions, in particular, should be made explicitly, by the board, in writing, and reviewed annually.
First: which categories of decisions in this institution will never be made by AI alone? A defensible Caribbean answer includes, at minimum, decisions that deny essential services, decisions that affect a person's legal status, decisions about children, decisions in mental-health crisis, and decisions affecting freedom of movement.
Second: what is the institution's posture on AI-generated content that interacts with the public? Will the institution disclose when a customer is interacting with AI? Will it disclose when a regulator, a journalist, or a court is reading material that was AI-drafted? The Caribbean cultural default, in most institutional contexts, is to expect the answer to be yes.
Third: what data will the institution allow to be exported to foreign AI vendors? Caribbean data protection laws and the underlying ethical concern about data sovereignty argue for a default of restriction, with explicit exceptions, rather than an open data flow with no policy. Board sign-off should be required for any new export of customer data to a foreign AI vendor.
Fourth: how will the institution handle requests from customers and regulators for an explanation of an AI-influenced decision? The institution that cannot answer this question in advance will, sooner or later, face a regulator or a court that requires it to answer in real time. Better to have the answer ready.
Fifth: what is the institution's posture on AI-generated content that misrepresents real persons (deepfakes of the institution's officers, AI imitations of customers' voices, synthetic identities)? Reactive policy after the first incident is more expensive than considered policy before it.
The Ethics of What Is Measured
Every AI system is shaped by what is measured. The choice of metric is the choice of values. Caribbean institutions deploying AI should pay particular attention to three measurement choices that quietly determine the ethical character of the deployment.
Accuracy is not a single number. Overall accuracy can be high while accuracy on a particular subgroup is unacceptably low. A Caribbean institution that reports only overall accuracy is choosing not to look at the subgroup data, which is also a choice. Subgroup performance, by sex, age, geography, ethnicity, and any other dimension on which there is real Caribbean variation, should be reported routinely.
False positives and false negatives are not symmetric in their human cost. A false positive in fraud detection inconveniences a real customer. A false negative may cost an institution money. A false positive in a child-welfare AI tool may shatter a family. A false negative may cost a child's life. Caribbean institutions should be explicit about which class of error matters more in each deployment, and design the tool to err in the chosen direction. The choice should not be left to the model's default settings.
What is not measured is invisible. An AI tool that does not measure waiting times, satisfaction with the human alternative, the number of customers who gave up on the process, or the experience of customers whose first language is not English, is producing a measurement of value that omits the dimensions most relevant to Caribbean public service. Caribbean institutions should add the Caribbean-specific metrics that the vendor's default dashboards omit.
Ethics Without Capacity Is Performance
There is a temptation, in Caribbean institutions, to respond to international ethics frameworks by producing a policy document, signing the international principles, and considering the work done. This is the failure mode the UNESCO Recommendation explicitly warned against. Ethics without operational capacity is performance. It produces an institution that has, on paper, all the right values, and in practice none of the controls that would let those values shape decisions.
Operational ethics capacity, for a Caribbean institution, has five components. A named officer with responsibility for AI ethics, with enough seniority to challenge a deployment. A standing place on the board agenda where AI ethics issues are tabled and recorded. An external party (a regulator, a regional body, a peer institution) that the officer can consult. A budget. A whistle-blower channel that has been used, at least once, without the user being punished.
The Caribbean institutions that have this combination of components produce sound AI ethics outcomes. The ones that have the policy and not the capacity produce avoidable ethics incidents and then defend them by pointing at the policy.
What Caribbean Citizens Can Demand
AI ethics is not a private institutional concern. It is a public good. Caribbean citizens have standing to demand specific things of the institutions, public and private, that use AI on them.
The right to be told when AI is being used in a decision about you. The right to ask for the basis of an AI-influenced decision. The right to a human review of a decision you disagree with. The right to know whether your personal data has been used to train a model that you are not a customer of. The right to expect that AI deployments which affect you have been considered, at the institutional level, with explicit attention to whether they serve Caribbean populations well.
These are not radical demands. Several of them are already legal entitlements in Caribbean jurisdictions with data protection legislation. The work in front of us is to make them lived realities, not theoretical rights.
Frequently Asked Questions
Are AI ethics frameworks just slow versions of regulation?
Not in the Caribbean today. Formal AI regulation is still emerging in most CARICOM jurisdictions. AI ethics frameworks are doing the load-bearing work that regulation will eventually take over. Caribbean institutions that build ethical practice now will find it easier to adapt to formal regulation when it arrives.
How is Caribbean AI ethics different from global AI ethics?
The global frameworks (OECD, UNESCO, IEEE, EU) converge on a similar set of principles. The Caribbean difference is in implementation: which subgroups need particular protection, which institutional structures provide the right oversight, which decisions are too sensitive to delegate, and which values worth coding in are missing from the global default. The principles travel; the implementation does not.
Should a Caribbean SME without specialist staff worry about AI ethics?
Yes, in a proportionate way. A small Caribbean business using AI in customer-facing decisions should be able to answer four questions: what AI is being used, on whom, with what oversight, and how a customer can ask for human review. That is the baseline. It is achievable without specialist staff.
Where can a Caribbean board start?
With the five board decisions listed above. They are concrete, they can be tabled at the next meeting, and they produce a written record that the board has taken AI ethics seriously. Everything else can be built on top.
What about the religious and indigenous ethical traditions across the Caribbean?
Caribbean public ethics is shaped by Christian, Hindu, Muslim, Rastafari, indigenous, and African-traditional traditions, and by the secular humanism of the region's universities. A Caribbean AI ethics worth the name cannot be reduced to any one of them. The four shared values identified above are an attempt to surface what is common across them, not a substitute for the deeper traditions.
A Caribbean Ethics Conclusion
AI ethics is too important to leave to the engineers who built the tools or to the international committees that wrote the frameworks. The Caribbean has its own ethical traditions, its own sense of what a good institution looks like, and its own experience of what happens when foreign technology is adopted without local ethical scrutiny. The work is to bring that experience to bear on the AI deployments now landing in our hospitals, banks, schools, and ministries, and to insist that the values being coded in are values we recognise.
Some of those values are universal. Most of them, in their specific Caribbean form, are not. That is not a deficit to be apologised for. It is a starting point.