In simple words: A label is an observed answer from the past, not a guess about the future.
Labels
A label is the known answer attached to an old example, such as Pass or Fail. Supervised learning uses these labelled examples to learn a relationship between inputs and answers.
Without labels
Unsupervised learning searches for structure without known answers, such as groups of similar customers. It can find useful clusters, but it does not automatically tell us what those clusters mean. Reinforcement learning uses a different setup: actions and reward signals.
See the flow
A table of students with attendance and final results can train a supervised pass predictor. A table without final results can only be grouped or analysed in other ways.
Remember: A label is an observed answer from the past, not a guess about the future.