Garbage In, Garbage Out

There's a famous saying in AI: "Garbage in, garbage out."

Feed an AI messy data and you'll get messy predictions — no matter how clever the AI is.

• Incorrect — wrong labels (a cat photo labelled 'dog')
• Missing — gaps where information should be
• Unbalanced — far more of one kind than another
• Duplicated — the same example repeated many times

Imagine training a cat-vs-dog AI on this data:

• A few cat photos are labelled "dog" → incorrect
• Some rows have no photo at all → missing
• 500 dog photos but only 4 cats → unbalanced
• One dog photo copied 30 times → duplicated

This AI will confidently call almost everything a dog. Not because it's dumb — because its data was garbage.

Here's a tiny dataset for a 'ripe / unripe' fruit AI:

• 20 ripe mangoes, 1 unripe mango
• One photo labelled "ripe" is actually unripe
• The same ripe photo pasted 5 times

Name every problem you can find.

Clean, correct, balanced data beats more data. A smaller tidy dataset often teaches better than a huge messy one.

How would you fix the unbalanced set of 500 dogs and 4 cats so the AI learns both fairly?