More than two hundred young people joined The Genius Project this year, the youngest aged five and the oldest eighteen, from across the Caribbean. Over one month we gave away 1M in cash and prizes, and no family paid tuition.
The decision people question most is the starting age. It is also the one I am most sure of, so this piece explains what five-year-olds do in the programme, what the older students built, how we trained parents, and the completion rate I would rather have been higher.
What five-year-olds do in the programme
They do not write code. Anyone who says they are teaching Python to five-year-olds is selling something.
A five-year-old in our programme sorts objects into groups and then argues about where the awkward object belongs. They count things and find that counting the same pile twice can give two answers if you are careless. They play a game where a machine guesses which card they are holding, and they find out that the machine can be confident and plainly wrong.
That last lesson matters most. A child who learns at five that a computer's answer can be wrong becomes a twelve-year-old who checks and a twenty-year-old who does not forward the fake. The habit of questioning has to form before the habit of accepting does.
By nine or ten the same ideas get names. Sorting becomes classification, the argument about the awkward object becomes a discussion of labels and edge cases, and the counting problem becomes measurement error. The vocabulary arrives after the intuition, which is the right order.
How the older students move from tools to mathematics
The programme starts with tools, because young people need to see a result quickly, and then moves to the mathematics underneath.
Teenagers went from prompting a model to understanding what a model is: training data, features, labels, and the difference between a system that learned a pattern and one that memorised the answer sheet. That leads to statistics: distributions, probability and why a single result tells you almost nothing. Then Python, notebooks and the ordinary experience of code that runs without error and returns the wrong number.
This is where most of our drop-off happens, and I come back to it below.
What the teams built
Teams worked on real problems instead of clean exercises, in four areas.
- Crime and community safety. Where incidents cluster, how gaps in reporting distort the data, and what a model can responsibly say about a place or a person. Several teams worked out for themselves that a model trained on incomplete crime data mostly predicts where reporting happens, a finding police forces in wealthier countries reached only after deploying such systems.
- Poverty and access. Household budgeting tools, food price tracking and matching people to services they qualify for but have not heard of. These teams hit the same wall professional teams hit: Caribbean household data is thin and scattered.
- Sport. Football and track data were the best starting point for machine learning we have found. Students already knew the sport, so they could tell at once when a model produced nonsense, which most beginners cannot do.
- Ethics and responsible AI. This ran through every project. Each team had to state who their system could fail, what data it should never hold and what they would say to a person the model got wrong. A fourteen-year-old explaining why her model should not be used for hiring decisions taught me more than many policy meetings I attend.
Every team built a working machine learning model: a system that took input, produced output and could be shown to be wrong.
A team that cannot say clearly what problem it is solving cannot build anything worth judging. That is true at fourteen and it is true in a boardroom. We just teach it earlier.
The parents' track
A child who understands AI better than every adult at home is exposed. So parents ran their own track alongside their children.
The practical part came first: account and privacy settings, what a chatbot keeps, what should never be pasted into one and how to spot a website built to harvest information. Then judgement: telling a generated image from a photograph, checking a claim before forwarding it, and recognising AI slop, the fluent and confident text that happens to be wrong. Fluent writing is often the reason people trust it.
Our target for parents was modest and specific: sit beside your child, look at the work and ask one question that improves it.
The final hackathon
The month ended with a hackathon in which teams presented to judges, defended their builds and answered for their choices. Congratulations to the winners and to everyone who presented. Defending a technical build to a panel of adults is hard at thirty, and several presenters were not yet thirteen.
The completion rate: about 15 percent
Completion across all programme areas currently stands at roughly 15 percent.
I publish it because a sector that reports only flattering numbers cannot improve. Large open online courses commonly report completion in the mid single digits, so 15 percent for a month-long technical programme across several countries is a result I will stand behind. It is also well below where it should be.
We lose people at the move from tools to mathematics, and we lose most of them where the internet connection is unreliable or there is nowhere quiet to work. Both can be fixed. Next year the mathematics runs in shorter modules, the materials work offline, and anyone who goes quiet for three days gets a phone call instead of a broadcast email.
Why the starting age matters for the region
The Caribbean cannot wait for AI literacy to come through the universities. The people who will build and govern these tools here in 2035 are in primary school today, and teaching responsible use now costs less and lasts longer than correcting habits at twenty-five.
In one version of the next decade, Caribbean young people operate tools built elsewhere. In another, they build and govern their own. What we teach children now, while their habits are forming, decides which one we get. Of everything I work on, this programme is the part I would keep last.
Who paid for it
The Genius Project charges families nothing, which works only because organisations across the region put in cash and support. For 2026 the supporters were:
- StarApple AI, which funded prize money and supplied instructors and curriculum;
- Maestro AI Labs, which gave technical mentorship and lab time for the machine learning tracks;
- the Caribbean AI Association, which backed the programme regionally;
- 14West, which supported the hackathon and the prize pool;
- AI Trinidad and Tobago, which carried it to students across the twin islands;
- Orbital Brand Science, which contributed in-kind support and helped reach families.
I put in my own money and taught in it, as I have every year.
To every sponsor, judge, volunteer instructor and parent who gave up a month of evenings, thank you. To the students, you did the hard part.
What to do next
- Parents of children aged five to eighteen: register for the next cohort at beagenius.org. It is tuition-free and needs no prior coding. Sign yourself up for the parents' track at the same time.
- Parents this week: point a phone's image recognition app at five odd objects around the house with your child and talk through every wrong answer it gives. That is the card game from our programme, done at home.
- Teachers: use one real local dataset, such as school sports results or rainfall records, and have students predict an outcome and then find where their prediction failed before introducing any code.
- Companies and foundations: contact the programme through beagenius.org to fund a specific barrier we identified, such as offline materials or follow-up calls for students who drop out at the mathematics stage.
- Other youth programme leaders: publish your completion rate alongside your enrolment figure, and state where in the programme people leave.
Frequently Asked Questions
Can a child under 13 use ChatGPT or Claude on their own?
Not under the providers' own rules. OpenAI's terms require ChatGPT users to be at least 13, with a parent's or guardian's permission under 18, and Anthropic requires users of its consumer Claude apps to be 18 or over. Younger children should use AI only in a supervised setting, such as a programme or classroom account set up by an adult, and parents should check each tool's age policy before creating an account.
What does a child need to take part?
No prior coding is required, and the programme is free. Because it runs online across several countries, a child needs a device and an internet connection for live sessions, and younger children need an adult nearby. From the next cohort the course materials will also work offline, which helps families with unreliable connections; check beagenius.org for the current requirements when you register.
How can parents tell whether a child is learning to question AI or just using it?
Ask your child to show you one answer an AI tool got wrong this week and how they found out. A child who is learning well can name the mistake, say what source they checked against and explain why the tool might have made it. A child who can only show you finished work the tool produced is using it, not questioning it.
What is AI slop, and how can families spot it?
AI slop is generated text, images or video produced in bulk with little care for accuracy, often written fluently enough to seem trustworthy. Warning signs include confident claims with no named source, oddly generic wording, images with distorted hands, text or backgrounds, and accounts that post large volumes on every trending topic. Before forwarding anything surprising, search for the claim on a known news site or the official body it mentions.
How does The Genius Project measure completion?
A participant counts as having completed only if they finish every programme area, including the mathematics and the final build, not just if they attend. That strict definition is one reason the figure is about 15 percent. We also track where each person stopped, which is how we know most leave at the move from tools to mathematics.