Benefits and limits

The idea

AI is fast and tireless, but it can be wrong or unfair, so people must check it and choose its data with care.

Good, bad or both?

Some people say AI is wonderful. Others say it is risky. What do you think?

• It is all good, with no problems
• It has real benefits and real limits, and it depends how we use it
• It is mostly dangerous and we should stay away

Weigh it up

Each card describes something about AI. Is it a benefit or a limit or risk?

Some cards are tricky: a benefit can turn into a risk. Tap a group. Read the reason, then press Next card.

Use it with open eyes

The benefits were speed, no tiredness, and more time for people. The limits were mistakes, missing or unfair data, and the need for human checking.

Neither side cancels the other. A good user keeps the benefits and watches for the limits, without fear and without blind trust.

AI is fast and tireless, but it can be wrong or unfair, so people must check it and choose its data with care.

Fair, varied and well-checked data make AI more reliable.

Notes

AI is fast and tireless, but it can be wrong or unfair, so people must check it and choose its data with care.

Think of it as a very useful tool: it needs a careful person behind it.

Check yourself

A school wants an AI tool to mark essays. Which plan gives the benefit and handles the limit?

A phone's face unlock works well for some family members and badly for others. What is the likely cause and best fix?

A district office wants an AI to warn about floods. What does it need most?

• Stop using it, since a tool can mark an essay wrongly. Giving up loses the benefit of saving teachers' time. A better plan is to use the tool carefully.
• Use it to save time, and let teachers check the marks — correct. Yes. The tool is fast, and people catch its mistakes and listen to students who disagree. That keeps both the benefit and the fairness.
• Use it for all the marks, because the tool is fast and checks itself. A tool cannot check its own marking. It may misjudge an essay written in an unusual way, so a person must look.
• It learned from few faces like theirs, so add more varied faces — correct. Yes. The machine did not see enough faces like theirs. More varied examples help it work for everyone.
• Those family members have faces that are too hard for any camera. Nothing is wrong with their faces. A tool should work for the people who use it, and its examples should include them.
• Face data cannot be improved, so some people have to be left out. The data can be improved. Fairness takes effort, but it is possible.
• A very fast computer that can work through data quickly. Speed alone does not help. Without records to learn from, a fast computer has nothing to find.
• A guarantee from the maker that the warnings will be correct. No AI can guarantee that. Its warnings are predictions, so people must stay alert and check.
• Many years of good records on rain, rivers and past floods — correct. Yes. AI learns patterns from data, so it needs plenty of good records to learn from.