AI Mistakes and Limitations
AI uses mathematical rules and data to make predictions. Missing or mistaken training examples and unclear inputs can contribute to errors. Explore mistakes and ways to check predictions.
What a learner can do afterwards
- Give an example of AI making a mistake, such as mishearing speech or changing typed text incorrectly
- Explain how missing examples or mistaken labels in training data can contribute to AI errors
- Describe how AI uses mathematical rules and data to make predictions
1 · Read
You have learned how a picture sorter trains on examples. AI uses mathematical rules and data to make predictions. A voice helper predicts words from sounds. Autocorrect can suggest changes to typed words. A recommendation tool can use earlier choices as clues. Each can make mistakes.
Imagine a sorter calling a muffin a puppy. Both might look brown and round in a photo. Missing training examples could also matter. A sorter shown mostly retrievers may struggle with poodles. These are possible explanations. A mistaken guess alone does not tell us exactly what caused it.
Missing examples and mistaken labels can teach misleading patterns. Blurry photos or unclear sounds can cause trouble too. Useful training examples can help, but do not promise perfect answers. Check guesses on new examples against known answers. Look for when mistakes happen before deciding what to change.
AI uses maths and data to predict. Training problems and unclear inputs can cause mistakes. Check guesses on new examples.
Lesson, questions and flashcards written for this dataset. The photo examples and test results are invented teaching examples.
These readings were independently written for this dataset. The school-library and picture-sorter situations are hypothetical.
More to read
Check the information behind a reply
A chatbot writes replies to messages. A clear reply can still contain a mistaken fact.
Imagine asking when your school library opens. The chatbot says eight o'clock. Your school's notice says nine. The chatbot might not have the notice or the right information for your school.
Ask a trusted adult to help check the school's notice. Compare its date and opening time with the reply. Sounding certain does not make the chatbot's answer reliable.
A chatbot can help draft a message. When an answer contains facts you need to use, check those facts with a suitable source.
Find when a sorter struggles
Imagine a sorter that labels hats and shoes. It calls a shoe a hat. Before changing its training examples, find out more.
Check that you know the photo's correct label. Try other new photos with different angles, light and backgrounds. Write down which guesses match their labels.
If the sorter often misses side views, check whether its training set includes useful side views. Also look for mistaken training labels. These checks investigate possible causes.
After making a change, test again on photos that were not used for training. Compare the results. Adding more copies of the same photo does not add the variety of different views.
2 · Watch
En anglaisTake it off screen
Where it sits
Where this leads
Jobs that lean on this skill. Follow one to see everything it is built on.
8 questions wait behind this lesson, each with its answer explained. Every answer feeds the sky: stars light as they are learned, and dim when it is time to come back.