Agentic Colonization: A Caribbean Agenda for Cultural Sovereignty

The Caribbean has a practical AI question to answer: when a system starts making choices on our behalf, whose interests travel with it? A tourism operator may welcome an assistant that handles bookings. A ministry may want faster research. An artist may need help reaching an audience. Each use can deliver value. Together, these choices can also move influence over regional commerce, public information and culture into systems whose priorities the region has little power to shape.

Adrian Dunkley calls this risk “agentic colonization” in his essay, “Agentic colonization and the decisions we give away”. His framework traces a progression from access, to integration, to decision influence. The term describes a potential dependency and power imbalance. It does not establish that every overseas AI service is exploitative, or that a particular company has committed an offence. Its value is the question it forces into procurement and policy: what authority are we handing over along with the work?

Follow the movement of authority

Access begins innocently. An organisation gives an assistant documents, customer queries or a calendar. Integration follows when the assistant becomes part of daily operations. Decision influence grows when people accept its shortlist, its explanation or its recommended action without an independent check. These stages can overlap. A small business can move through all three before it has written a policy.

Consider a hypothetical regional travel assistant. It might favour suppliers that are easy to book through its integrations. An independent guesthouse could disappear from the shortlist even when it offers a better cultural experience. That outcome need not involve deliberate hostility. Technical access, ranking criteria and commercial arrangements can still determine which businesses become visible. A procurement team should ask how those choices work before making the assistant its default gateway.

Culture is part of the infrastructure

Cultural sovereignty includes the ability to decide how a community’s language, histories and creative work are represented and used. It cannot be reduced to placing a regional flag on a chatbot. A system may reproduce Caribbean imagery while misunderstanding a place name, flattening a dialect or presenting a living tradition as a tourist attraction.

The risk of cultural dilution grows when the same generated conventions are repeated across school materials, advertising, search summaries and entertainment. Familiarity can begin to look like authenticity. Communities then spend time correcting a version of themselves that automated distribution has made convenient. These are risks to examine, not evidence that every generated description will be wrong.

A useful regional evaluation would ask local speakers, historians and practitioners to test real tasks. Can the system preserve meaning across different language varieties? Does it distinguish neighbouring territories? Does it acknowledge uncertainty about a tradition? Who can challenge an error, and how quickly does the correction reach users? Cultural expertise belongs in the evaluation budget.

A UNESCO-convened expert assessment of AI and culture similarly raises concerns about platform dominance, cultural homogenisation and creative agency. That is a policy assessment of risks and priorities, rather than proof that any particular service has harmed Caribbean culture.

Ask who captures the creative value

A difficult economic pattern deserves public scrutiny. Creative work can help make a platform useful; the same platform may later charge creators for tools, reach or access to audiences. When creators become dependent on that intermediary, bargaining power can shift away from them. This is the extraction-and-resale concern at the heart of the argument.

The details matter. Training, licensing, copyright, distribution fees and advertising are separate questions. A claim that particular work was used unlawfully requires evidence about that work, the relevant use and the applicable law. Regional policy should demand transparency and fair terms without presenting every disputed use as proven theft.

Practical responses include clear licensing options, records of consent, meaningful attribution where appropriate and negotiated terms for collective cultural collections. Public institutions should be especially careful with archives. Possessing a digital copy does not automatically settle whether every person or community represented has agreed to every commercial reuse.

Media influence needs a public-interest test

An agent that summarises the news becomes an editor of attention. It chooses which stories appear, how competing accounts are described and which sources users can easily open. Personalised persuasion could amplify manipulation if the system combines detailed knowledge of a person with an undisclosed commercial or political objective. That is a credible risk scenario, not a claim that every assistant is running such a campaign.

Regional broadcasters and newsrooms should test whether summaries preserve attribution, distinguish reporting from opinion and link readers back to original journalism. Institutions using agents for public information should publish the system’s role and keep accessible alternatives. Citizens need a way to see the underlying source, question the summary and reach a responsible human.

Put safeguards into the purchase

A shared Caribbean procurement checklist could turn principles into bargaining power. Before adopting an agent, an institution should establish five things:

• The exact tasks and data the system may access, with sensitive material excluded unless its use is justified and authorised.
• The actions that require human approval, especially public statements, consequential decisions and spending.
• How recommendations are ranked, including paid placement and preferred integrations where these apply.
• How records, content and workflows can be exported, and what it costs to leave.
• Who investigates a complaint, corrects a cultural error and takes responsibility for an unauthorised action.

Answers should be tested in a small pilot using realistic cases, then checked again when capabilities or terms change. A vendor’s promise is weaker than a demonstrated export, a working approval control or an auditable record of an action.

Build regional capacity together

No single island needs to duplicate every technology stack. Universities, libraries, creative associations and public agencies could share evaluation methods, training materials and carefully governed language resources. Regional cooperation can also support independent technical review and stronger negotiation over terms.

The aim should be practical freedom of choice. Local development can strengthen that freedom, while imported tools can remain useful when their conditions are acceptable. A local label alone does not guarantee accountability; ownership, governance and incentives still deserve inspection.

Agentic colonization gives the Caribbean a vocabulary for discussing power before dependence becomes difficult to reverse. The next step is measurable: make one procurement checklist public, test one consequential workflow and involve the people whose culture the system describes. Technology serves the region best when the region retains the capacity to question it, correct it and walk away.