01 What is AI?
Artificial because it isn’t human: it runs on computers. Intelligent because it can do four things:
1. Learn
Pick up patterns from examples and data.
2. Reason
Work through information logically.
3. Solve problems
Find answers to new questions.
4. Make decisions
Choose what to do next based on what it learned.
Wanjiku opened a bakery in January. Sales peaked in February (Valentine’s), were quiet in March, rose in August (wedding season), and picked up again in December (holidays, weddings and graduations).
Next year she won’t guess. She’ll look at last year’s sales and stock up before February, August and December. That is what AI does: it looks at past data, learns the patterns, and makes decisions.
The difference? Wanjiku has one year of sales data. AI has been exposed to enormous amounts of data, so it can spot far more patterns.
In one sentence: artificial intelligence is computer software that has been exposed to data and has learned patterns from it.
02 The types of AI
AI can be grouped in two ways: by what it can do (functionality) and by what it produces (output).
By functionality
| Type | What it means | Where we are |
|---|---|---|
| Narrow AI | Good at one specific task: writing an email, recommending a song, translating English to Kiswahili. | This is the AI most of us use today. |
| General AI (AGI) | AI that can reason, think and adapt like a human across many different areas. | Not there yet, but early versions are appearing. |
| Super AI | AI that is more intelligent than humans. | Still theoretical: the AI of movies. |
By output
Predictive AI
Good at guessing what happens next, like a weather forecast or a sales prediction.
Generative AI
Creates new things: text, images, videos and songs. Most of us use this every day.
03 AI models
Think of a model as a student. Students go to different schools (village, town, abroad) and study similar subjects with different teachers and resources. Each one learns slightly differently.
AI models are the same. ChatGPT, Claude, Gemini, Grok and Copilot each went through training, a period of learning. When you ask a model a question, it infers an answer from what it learned, just as you answer from what you learned in school.
When a company releases a new model, it is like that student completing a master’s degree: they come back with more knowledge.
Watch out for hallucination. If someone asked you about Pythagoras’ theorem today, you might half-remember it and confidently give a wrong answer. AI models do the same. That is called hallucination.
What to do: always fact-check AI answers, ask it to iterate, and compare a few variations before you rely on the output.
04 AI and data
Think of a DJ. To perform, a DJ needs three things, and so does AI:
| A DJ needs… | AI needs… |
|---|---|
| Music tracks | Data: text, images and video to learn from. |
| A mixer | An algorithm that processes the data. |
| An audience | You, the user. |
Both the quantity and the quality of the data matter. Because AI depends on data, watch for four issues:
- Bias: a DJ who only listens to gospel can’t play an amapiano concert. AI trained on narrow data gives narrow, biased answers.
- Garbage in, garbage out: a DJ who knows three songs shouldn’t headline a concert. Poor data means poor results.
- Prompting: “play at a gospel show” gets a generic set. “Play a gospel show for 25 to 30 year-olds” gets the right one. The more context you give AI, the better the answer.
- Data privacy: anything you paste in could shape what the model says to others.
Data privacy example: you ask AI to draft a thank-you email to Kevin, your best client, who buys 20 pairs of shoes a month and always pays on time. Later, someone else asks the AI to find a client who buys more than five pairs a month. Who might it point them to? Kevin.
Before you paste anything into AI, ask: could this data hurt my company or cost us a client?
05 Practical uses of AI
Grow your business
Describe what you sell and to whom. Ask for a social media content strategy, ideas for approaching clients, and pricing ideas.
Plan your health goals
Share your current and ideal weight and your work schedule. Ask for practical meal plans that fit your day.
Learn anything
Ask AI to teach you a topic step by step, such as how money market funds (MMFs) work and how to invest in them.
06 AI jargon, decoded
| Term | Plain-English meaning |
|---|---|
| LLM | Large Language Model. AI trained on huge amounts of text. ChatGPT and Claude run on LLMs. |
| AI agent | Goes beyond a chatbot. A chatbot suggests places to stay in Mombasa. An agent also checks bookings and flight prices, advises on the best time to go, and books the flight, like an employee who does the task from start to finish. |
| Prompt | How you talk to AI. The more descriptive you are, the better the answer. |
| MCP | Model Context Protocol. Like a universal charger with a cable for your phone, earphones and laptop, it lets AI connect to many tools at once, such as your Gmail and calendar. |