What AI Is, Where It Is Used, and How to Use It Responsibly

1AI is already around you

Since October 2016, drones run by the company Zipline have flown blood to hospitals across Rwanda, first from a centre in Muhanga and, since 2018, also from Kayonza. The drones navigate and avoid obstacles on their own, using software instead of a pilot. A 2022 study found that the service cut blood delivery times by 61% and reduced the number of blood units that expired unused by 67%.

Closer to home, your phone keyboard guesses your next word, a translation app turns English into Kinyarwanda, and a video app decides what to show you next. All of these use artificial intelligence (AI). This unit teaches you what AI is, how to use AI tools well, and how to use them responsibly.

Worked example: following rules versus learning

A program that follows rules. A programmer writes: "if the temperature is above 30 °C, show 'Hot'". The computer follows that rule exactly and never changes it.

A program that learns. A spam filter is shown thousands of emails that people have already marked "spam" or "not spam". It finds patterns on its own, such as certain words, links and senders, and uses them to sort new emails it has never seen. Nobody wrote a rule for each pattern; the program learned them from examples.

2What is AI?

Artificial intelligence is the ability of computer systems to do tasks that usually need human intelligence, such as learning, reasoning, solving problems, recognising speech or images, and making decisions. Most modern AI uses machine learning: the system learns patterns from large amounts of examples, called training data, instead of following only rules written by a programmer. What the system produces from its training is called a model.

Worked example: training a model

A fruit cooperative wants a phone app that sorts mangoes into "ripe" and "not ripe" from a photo.

1. Collect data: photograph 1,000 mangoes and label each photo "ripe" or "not ripe".
2. Train: the model finds the patterns that separate the two labels, such as skin colour and softness marks.
3. Test: show it 200 new photos it has never seen. It labels 180 correctly: 180 ÷ 200 = 0.9, an accuracy of 90%.
4. Use and monitor: put it in the app and keep checking its mistakes.

Now suppose every training photo showed a variety of mango that turns yellow when ripe. Shown a variety that stays green when ripe, the model will say "not ripe" every time. A model is only as good as its training data.

3Where AI is used

4Kinyarwanda and AI

An AI system can only understand a language it has been trained on, and for a long time very little recorded Kinyarwanda speech was available. Volunteers have changed that. Mozilla's Common Voice is an open project in which volunteers record and check sentences in their own language, and the recordings are given away free for anyone to use. Kinyarwanda is now one of the five languages with the most recorded hours on it. In Kigali, the company Digital Umuganda collects voice, text and translation data for African languages and builds tools on that data, such as its Mbaza chatbot. Open data like this lets developers build speech tools for Kinyarwanda speakers.

5Using AI responsibly

• Accuracy: AI can be confidently wrong. Check important facts in a trusted source.
• Bias: an AI system can treat some groups of people unfairly when its training data does not represent everyone. This is called algorithmic bias. It can come from what data is collected, how it is labelled, or where the system is used.
• Privacy: what you type or upload may be stored and used to train future models. Never enter passwords, PINs, ID numbers or phone numbers, and never upload photos of other people without their consent. Rwanda's 2021 law on personal data and privacy protects people's personal information.
• Honesty: presenting AI-made work as entirely your own is cheating. Follow your school's rules and say when you used AI.
• Human responsibility: a person must stay accountable for decisions that affect people's lives, such as marks, medical care or jobs.
• Environment: training and running large AI models uses a great deal of electricity, and water for cooling data centres.

In 2023 Rwanda adopted a National Artificial Intelligence Policy to guide how AI is developed and used in the country.

Worked example: spotting bias

A speech-recognition system is trained mostly on recordings of adult men from Kigali. It is tested on 100 sentences from each of three groups. It gets 94 right for adult men from Kigali, 81 for women, and 72 for children from rural districts. The gap between the best and worst groups is 94 − 72 = 22 percentage points. The cause is the training data: the system rarely heard voices like those of rural children. The fix is to collect recordings from many kinds of speakers, which is exactly why Common Voice asks for many different voices and accents.

Practice