When the Machine Is Wrong
Has a machine ever done something silly? Maybe a voice assistant misheard your question, or a spell-checker changed your word to something totally wrong. If that has ever frustrated you, you are not alone. Machines make mistakes — and that is completely, one hundred percent normal. In this lesson we are going to find out WHY machines get things wrong, and — here is the exciting part — what we do about it.
Why Machines Make Mistakes
Remember how a machine builds a model by studying examples? That model is its best guess at the pattern. But no model is perfect. There are three common reasons a machine makes a mistake: 1. It never saw an example like this one. If a machine trained only on sunny-day photos of dogs and you show it a dog in a raincoat, it might be confused — it has no raincoat-dog experience! 2. Its examples were biased. If most of the examples leaned one way, the machine leans that way too. 3. The new example is genuinely tricky. Sometimes a cat sits in a weird position that looks a bit like a dog. Even a person might double-check! When a machine makes a wrong guess, scientists call that an error or a mistake. Getting errors does NOT mean the machine is broken. It means there is room to learn more.
Mistakes are how we know what the machine has not learned yet. Every mistake points to a gap — and we can fill that gap by giving the machine better examples.
Here is a story. Alex trained a machine to sort mail. It learned to recognize "letter" versus "package." It did great on normal envelopes and boxes. Then one day, a very large, puffy envelope arrived — more like a small pillow. The machine said "letter." Wrong! It had never seen a padded envelope before. Did Alex throw the machine away? No! Alex gathered ten photos of padded envelopes, labeled them "package," and added them to the training set. The machine trained again, and the next puffy envelope — sorted correctly. The mistake was useful information. It told Alex exactly what example was missing.
Match each mistake cause to the right fix.
Terms
Definitions
Drag terms onto their definitions, or click a term then click a definition to match.
It is also important to know that sometimes a machine will be wrong and very confident about it. That is extra important to watch out for. Imagine a friend who guesses every answer with total confidence, but half the time they are wrong. You might prefer a friend who says "I am not sure about this one" when they really are not sure. Good AI designers pay close attention to mistakes AND to confidence. They do not just count how many times the machine is right — they also look carefully at the cases where it is wrong.
Machines can be wrong — and sometimes very confidently wrong. If an AI is helping make an important decision, a person should always check the answer. AI is a helper, not the final word.
Alex's machine misidentified a padded envelope. What is the BEST way to fix this?
Why can a machine be very confident AND very wrong at the same time?
Mistake Detective
- Think of a time a machine gave you a wrong or silly answer — a voice assistant, spell-checker, autocorrect, or a phone camera that labeled the wrong thing.
- Write down or draw what happened.
- Now be a detective: what KIND of mistake was it? Did the machine never see this situation before? Was it a tricky case?
- If you were the machine's trainer, what examples would you add to help it do better?
- Share your investigation with someone at home.