Course 1 - Module 1 - Lesson 2
The helpful helper test: when AI is useful, unsure, or plain wrong
Meet the AI Toolbox
Judge the task before the answer
AI is often useful when there are many acceptable answers and a person can easily judge them. Brainstorming names, rewriting a clear message, or organizing supplied notes are good practice tasks. You can compare the output directly with what you wanted.
It becomes harder to rely on an answer when the task depends on missing context, a recent change, or a precise fact. A confident answer about tomorrow's opening hours may simply be a plausible guess unless the application checked a current source.
Use the helpful helper test
Ask three questions: Do I have enough information to describe the task? Can I check the result? What happens if it is wrong? A lunch suggestion is easy to replace. An incorrect payment instruction can cause a much larger problem.
For example, ask an assistant to shorten a volunteer announcement you provide. Check that it preserves the meeting place and time. If you instead ask it to invent the meeting details, the polished announcement may contain information that nobody agreed to.
Make uncertainty visible
Ask the assistant to separate supplied facts, assumptions, and unanswered questions. This makes review easier, but the assistant's own confidence is not independent evidence. Verify important statements against an original document, a reliable source, or someone responsible for the decision.
When checking costs, recalculate the total. When checking a quotation, open the original passage. When essential information is unavailable, leave the field blank or ask for it. A useful helper should make a missing detail easier to notice.
Tiny experiment
Sort these tasks into easy to check or needs outside evidence: shorten your note, confirm tomorrow's train time, and suggest three party themes. Explain one choice.
Confidence check
What is one familiar task where a believable answer could be harder to check than you first expect?