What is AI? A plain guide for Jamaicans

If you own a phone in Jamaica, you have used artificial intelligence today. Probably before breakfast.

It sorted your email so the scam about a “lottery win” went to spam. It picked the route around the traffic on your way to work. It decided which video showed up first when you opened TikTok or Instagram. None of that felt like the future. It just felt like your phone.

So why does everyone suddenly talk about AI like it arrived last year? Because a new kind arrived. One that writes, talks and answers back. That’s the part people mean now, and it’s the part this guide is mostly about.

We build AI assistants for businesses in Jamaica, so we spend our days with this stuff. Here’s what it is, in plain words.

The short answer

Artificial intelligence is software that learns from examples instead of following rules a person typed out one by one.

Old software works like a recipe. Somebody writes every step: if the customer types “1”, show the price list. If they type “2”, show the address. The computer never does anything the recipe didn’t say.

AI works more like how a child learns what a mango is. Nobody gives a child a rulebook about mangoes. They see hundreds of them, ripe and green, East Indian and Julie, and one day they just know. AI is trained the same way, on huge numbers of examples, until it can recognise the pattern by itself.

That’s the whole idea. Everything else is detail.

The AI you already use

Long before chatbots, AI was quietly running inside the apps you use every day:

• Spam filters. Your email learned what scams look like from millions of examples, so you rarely see them.
• Maps. Route apps learn traffic patterns and predict how long each road will take right now.
• Your bank. When the bank texts to ask if that charge was really you, a system flagged it because it didn’t match your normal spending.
• Social media feeds. The order of every post you see is chosen by a system that learned what keeps you watching.
• Your camera. Portrait mode, the face scan that opens your phone, and the filter that smooths your skin on a video call all come from systems trained on photos.

This kind of AI is narrow. It does one job well and nothing else. The spam filter can’t write a letter. The map can’t tell you a joke.

The new kind: AI that talks back

The AI everyone talks about now is called generative AI, because it generates things: text, pictures, voices, code.

The best known is ChatGPT, made by a company called OpenAI. In October 2025 its boss said ChatGPT had passed 800 million weekly users. There are others too: Google’s Gemini, Anthropic’s Claude, Microsoft’s Copilot, and Meta AI, which is built into WhatsApp and Instagram in many countries.

The thing inside them is called a large language model. “Large” because it learned from an enormous amount of text: books, websites, articles, forums. “Language model” because what it learned is how language fits together.

How it actually works, without the maths

Here’s the surprising part. At its core, a language model is doing one thing: guessing the next word.

Give it “Good morning, how are” and it predicts “you”. Give it a question about your light bill and it predicts, word by word, what a helpful answer would sound like. It has seen so much writing that its guesses come out as full, sensible paragraphs.

It learns in two stages:

1. Training. The company feeds the model a mountain of text and lets it practise guessing the next word, billions of times, adjusting itself each time it guesses wrong. This takes months and huge data centres.
2. Using. Once trained, the model is frozen. When you type a question, it isn’t learning from you in that moment. It’s using what it learned in training to predict an answer.

If you want a beautiful visual version of this, the Financial Times made one: Generative AI exists because of the transformer. If you want a proper free course with no maths, try Elements of AI from the University of Helsinki.

What AI is good at

Used well, these tools save real time. They’re good at:

• First drafts. A letter to a landlord, a job application, a price list, a Facebook post.
• Summaries. Paste in a long document and ask for the five points that matter.
• Explaining things. Ask it to explain a contract clause, a school maths problem or a medical term in simple words. Then check it.
• Answering the same question many times. This is where businesses use it most. A shop that gets asked “What time you open?” and “How much for delivery?” all day can let an assistant answer instantly, at 2 AM, without getting tired.

That last one is what we build. Our AI front desk answers a business’s calls and WhatsApp messages, takes the customer’s details and books the job. You can see the ones we’ve deployed for a shipping company and a medical clinic on our work page.

What it gets wrong

This is the part the advertisements skip.

It makes things up. AI can state something false with full confidence. The industry calls this a “hallucination”. OpenAI’s own researchers published a paper in September 2025 arguing that models do it partly because of how they’re trained and tested: guessing gets rewarded more than saying “I don’t know”. So never trust a fact, a law, a price or a medical answer from AI without checking a real source.

It doesn’t know what happened this morning. A model knows what was in its training data. Unless the tool is searching the web for you, recent news is a blank.

It speaks better English than Patois. These models learn from written text, and there is far less Patois written online than standard English. Patois is the everyday language of most Jamaicans, yet the first dataset for testing how well AI understands it was only published in 2022, by Ruth-Ann Armstrong, John Hewitt and Christopher Manning. Expect the tools to handle “Mi soon come” worse than “I’ll be there soon”.

It reflects what it learned. If the training data leans one way, so does the AI. It learned mostly from the internet, and the internet is mostly American and European. Ask it about Jamaican law, Jamaican prices or Jamaican places and it may confidently give you an American answer.

AI in Jamaica right now

Jamaica isn’t just watching from the side.

• The government has a task force. The National AI Task Force was set up under the Office of the Prime Minister in 2023. Its policy recommendations were published in February 2025, covering everything from economic growth to education and training workers.
• A national AI policy is on the way. In February 2026 the task force was reconstituted with 23 members and asked to deliver a draft national AI policy within six to nine months, which made it due around November 2026.
• Almost everyone is online. About 2.54 million people in Jamaica used the internet at the end of 2025, close to 9 in 10 people. That’s the audience every AI tool, and every business using one, is now reaching.
• Businesses are starting to use it. Mostly in small ways so far: AI-written social posts, chat assistants on WhatsApp, tools that answer the phone. This is the area we work in every day.
• Jobs are already moving. Our call centre sector has shrunk in the past two years. The industry association told The Gleaner that AI isn’t one of the causes. We’d add the word “yet”, and that’s worth a whole article of its own.

Words you’ll hear

• Model. The trained system itself. ChatGPT is an app. The model inside it has its own name, like GPT-5.
• Chatbot. Any program you chat with. Old ones followed scripts. New ones run on language models.
• Prompt. What you type to the AI. Better prompts get better answers.
• Generative AI. AI that makes new text, images, audio or video.
• Hallucination. When the AI states something false as if it were true.
• AI agent. An AI that doesn’t just answer, but takes actions: books the appointment, sends the email, fills the form.
• Physical AI. AI with a body. Robots, self-driving cars, machines in warehouses and factories.

How to start using it today

You don’t need a course to begin. Try this:

1. Pick one free tool. ChatGPT, Gemini, Claude and Copilot all have free versions.
2. Give it a real job. “Help me write a polite message to a customer who hasn’t paid in 30 days.” “Explain this paragraph from my insurance policy like I’m 15.” “Give me 10 names for a cook shop in May Pen.”
3. Talk to it like a person. If the answer is wrong or too long, say so. “Shorter.” “Less formal.” “That’s not true in Jamaica, try again.”
4. Check anything that matters. Laws, prices, medicine, money. Always confirm with a real source.
5. Keep private things private. Don’t paste your TRN, passwords, bank details or anyone else’s personal information into a chatbot.

The bottom line

AI isn’t magic, and it isn’t going away. It’s a very good guesser that learned from more text than any person could read in a thousand lifetimes. That makes it fast, useful and sometimes confidently wrong.

The people who do well with it won’t be the ones who know the most about how it works. They’ll be the ones who figure out, early, which parts of their own work it can take off their hands.

Sources

1. TechCrunch, Sam Altman says ChatGPT has hit 800M weekly active users, October 2025.
2. OpenAI, Why language models hallucinate, September 2025.
3. Armstrong, Hewitt and Manning, JamPatoisNLI: A Jamaican Patois Natural Language Inference Dataset, Findings of EMNLP 2022.
4. Office of the Prime Minister, National Artificial Intelligence Task Force Policy Recommendations, February 2025.
5. Office of the Prime Minister, first meeting of the reconstituted National AI Task Force, 2026.
6. DataReportal, Digital 2026: Jamaica.
7. The Gleaner, report on outsourcing job losses, with experts on the role of AI, July 2026.
8. Financial Times, Generative AI exists because of the transformer.
9. University of Helsinki and MinnaLearn, Elements of AI.