AI is showing up in Caribbean governments, schools, and businesses faster than the policy conversation is moving. From Guyana, the ethics questions look different than they do in a Silicon Valley keynote: data ownership, who benefits, and what happens when tools trained elsewhere get dropped into local institutions.
What is actually at stake
In a small state, "AI ethics" is not an abstract seminar. It collides with budgets, power cuts, and the fact that most of the models were not built for us.
Data. If a ministry or a bank sends records to a vendor in the North, the value of that data leaves with it. We have seen this pattern with other resources. Calling it "the cloud" does not change the direction of the money.
Bias. Models trained on Western faces, English, and Western institutions will misread Caribbean faces, Creole, and how people here actually talk to a counter. Facial recognition, credit scoring, and chatbots all inherit that.
Jobs. Automation hits harder where the work is already thin: agriculture, tourism, clerical government jobs. Retraining programmes that exist only as a slide are not a policy.
Power and waste. Training and inference cost electricity. Hardware becomes e-waste. The Caribbean already lives with climate and import bills. Pretending AI is weightless is a lie.
Agriculture is a useful test
Rice on the coast is the example I keep coming back to. A model that forecasts yield or flags blast disease could help. It could also push small farmers toward a vendor they cannot leave, or treat traditional knowledge as noise. Food sovereignty is the test: does the tool stay here, or does it make the farm dependent on someone else's API?
What I will not do
I will not write a CARICOM ethics checklist and call it a contribution. Shared guidelines, joint research, and coordinated policy are useful. They are also easy to announce and hard to staff.
I will not pretend "participatory workshops" and "local advisory boards" are a method if they are only a paragraph. If a system is going into a ministry, the people who will live with it should see it before it ships. That is the whole point. The rest is vocabulary.
What I will do
Argue the narrow questions in public: who owns the data, who gets the jobs, and who pays when a model is wrong. Build systems in Guyana. Prefer a disagreement to another general statement about ethical AI.
Caribbean states cannot copy a US or EU ethics checklist and call it done. I do not have a framework to sell. I have a few years of building here.
If you work on this, write back.