Information gain helps an agent choose among possible questions. A high-value question separates plausible interpretations or removes an important blocker, while a low-value question adds detail that is unlikely to change the outcome.
The best question is not always the one that gathers the most data. A practical agent balances expected uncertainty reduction against user effort, privacy, time, and whether the answer can actually change the next action.
ELI5
Information gain is the amount a new answer helps narrow down what is still unknown. An agent can use it to ask the question most likely to change or improve the decision.
For example, when choosing a child car seat, asking the child's weight may remove many unsuitable options immediately. Asking about a preferred shade of blue may add less useful information at that stage.
