In 2018 I started running free AI training sessions for students in Jamaica: every week, no charge, no prerequisites beyond turning up and taking part. Seven years later I am still doing it. The format, the students and the field have all changed, and I have now spent more hours teaching AI to beginners than on almost anything else. This piece sets down what that has taught me about how people learn AI, where commercial training goes wrong, and how you can run or find something similar.
Why the sessions started
At StarApple AI I kept running into a problem that was not about technology. The tools to do useful things with AI existed and were getting cheaper. What the Caribbean lacked was people who could spot where AI would help them, describe a problem clearly enough for a builder to work on, and judge whether the result worked.
That is a different skill from programming. It resembles being a good client on a construction project: you do not lay bricks, but you can describe what you need, read the plans and see when something is going wrong. Almost nobody in the region had the AI version of that skill, and the formal education system was not going to produce it quickly.
The first cohort was eight people in a rented room in Kingston: university students, working professionals and a secondary school teacher who heard about it from a friend. None had any AI background. Many sessions moved online during the pandemic and stayed there, because online brought in people from across Jamaica and sometimes other Caribbean countries. Some weeks two people come. Some weeks forty.
How a session runs
The sessions are workshops built around a real problem, usually one that StarApple AI or a Caribbean organisation I work with is trying to solve. I tried lectures in the early months and dropped them.
We work through the same questions each time:
- What do we know about the problem?
- What information would help us understand it better?
- How might an AI system approach it?
- How could that approach fail?
- How would we test whether it worked?
There is no code in anyone's first session. When I started with code, the room split within fifteen minutes into people comfortable with code and spectators. The skill I am teaching does not depend on programming, and starting with code confuses the two. Code comes later, once participants have already asked the question it answers.
The most important thing I teach is how to describe a problem well enough that an AI system could help with it. That is harder than most people think.
What I have learned about how people learn AI
These are observations from watching thousands of people meet AI for the first time. They are field notes from my sessions and have not been tested in controlled research.
The first barrier is fear of looking foolish
Most newcomers have heard enough about AI to believe that not understanding it says something bad about their intelligence. So they perform engagement: nodding at points that did not land and asking questions that sound clever in place of the ones that would clear up their confusion.
The fix is to make not understanding visibly safe. I ask questions I cannot answer, invite correction and treat the session as a shared investigation. Once people believe that admitting confusion is safe, they start learning.
Analogies build understanding faster than maths
For non-technical learners, the mathematics of machine learning transfers poorly to practical understanding. Analogies to familiar human situations transfer well:
- Training a model is like giving a new hire examples of good and bad work so they can calibrate their judgement.
- Fine-tuning is like giving an experienced hire on-the-job experience in your particular business.
- Overfitting is like a student who memorised the past papers and is thrown by a slightly different question.
None of these is technically precise. Each builds an accurate intuition faster than a formal explanation does.
Caribbean learners bring intuitions outside educators miss
Outside educators often describe Caribbean students with no AI background as "behind" their North American or European peers. I disagree with that description more strongly than with anything else in this piece.
People who grew up handling the informal economic complexity of Caribbean life often have strong intuitions for fraud detection, social networks, allocating scarce resources and deciding under uncertainty. They learned those skills in places where money is tight and trust has to be earned. What they usually lack is the vocabulary that links what they know to the formal language of AI. Once that link is made, they learn fast.
Where commercial AI training goes wrong
Every year, people who took online courses, bootcamps or corporate programmes arrive at my sessions still unsure of the basics. The failures I see most often:
AI literacy is taught as machine learning engineering. Most AI courses teach people to build and train models. That is a valuable skill, but a manager, teacher or analyst mostly needs to know what AI can and cannot do, how to judge a proposed system and how to talk to the people building it.
No domain focus. A doctor, a teacher, a financial analyst and a logistics manager need AI applied to different problems, with different risks and failure modes. Generic courses produce graduates who can describe AI in general but cannot name the places in their own week where it would help.
The certificate conveyor belt. Online courses reward completion, completion rewards speed, and speed rewards spotting the right quiz answer. People leave with a certificate and confident vocabulary, and the gap shows as soon as they join a real project.
What seven years of sessions produced
Some people from the early cohorts now run AI projects in Caribbean organisations, as the informed client: they define the problem, evaluate the solution and manage the relationship with the system and its builders. That was the missing capability when I started.
Others went on to formal AI degrees and could engage with the technical material from the first week. Some started businesses. Some now teach AI themselves, which matters more to me than any single career. I have no clean count of how many participants went on to AI work, and I would like one; that is the weakest part of my own evidence.
Why I keep doing it
People ask why I still give this time away when it has become more valuable commercially. There are two reasons.
Teaching keeps me honest. Every session makes me explain something clearly enough for a beginner to follow, which exposes the difference between what I understand and what I can merely talk about. It has made me a better practitioner.
The second reason is personal. I grew up with neurodivergence and learning disabilities in a school system that was not built for how my mind works. Teachers who found other ways to explain things, and family who made room for how I learned, changed my life. The sessions are the simplest way I have of paying that forward: turning up every week for anyone who wants to learn, at no cost, and teaching the way I wish I had been taught.
What to do next
- If you want to learn: write down one problem from your own work in two or three sentences before you pick any course. Choose training that lets you work on that problem, and use the contact page on this site to ask about the current session schedule.
- If you teach or run a community group: run one 90-minute session using the five questions above on a local problem, such as a school's attendance records or a church's event sign-ups. No code, no slides beyond the problem statement.
- If you manage a team: before paying for AI training, ask the provider to show you one exercise built on your sector's problems. If they cannot, keep looking.
- If you build training courses: replace one quiz module with a project where learners define a problem, propose an AI approach and list its failure modes, and grade that.
- If you already understand AI: commit to one free hour a month teaching someone in your field. Start with the analogies and leave the maths for later.
Frequently asked questions
Do I need to know how to code to start learning AI?
No. You can learn to frame problems, use AI tools well and judge AI output without writing code, and that covers what most professionals need. Coding becomes useful once you want to build or automate things yourself. Python is the usual first language, and free courses are enough to get started.
Which free courses are worth taking after a first session?
Elements of AI, from the University of Helsinki, is a free online course for non-programmers that explains the core ideas without maths. Andrew Ng's AI for Everyone on Coursera covers the same ground from a business angle; check whether a free audit option is offered when you enrol. For people who can already code, fast.ai's Practical Deep Learning course is free and project-based.
How do I tell whether a paid AI course is worth the money?
Ask to see a sample project and check whether it uses problems from your field or only generic examples. Find out whether instructors have built or run real AI systems, and whether you get feedback on your own work or only automated quizzes. A certificate on its own tells an employer little, so choose the course that leaves you with a project you can show.
How can I run a free AI session in my school, church or workplace?
Pick one real, local problem, such as scheduling, attendance or customer questions, and write it up in a paragraph. Spend the session on what you know about the problem, how an AI tool might help, how it could fail and how you would test it, then try one free AI assistant on it together. Keep the group under fifteen for discussion, and make it clear from the start that questions are welcome.
Is it safe to put my work data into free AI tools while practising?
Not unless your employer allows it. Free versions of AI assistants may use conversations to improve their models unless you switch that off in the settings, and anything you type is processed on the provider's servers. Practise with made-up or public data, and ask your employer which tools are approved for real work.