Why the question matters
Socrates is known largely through the writings of others, including Plato. In those dialogues, questioning often unsettles an answer that initially sounded certain. That tradition offers a useful lens for thinking about AI at work, without settling every modern question for us.
This is an interpretive essay, not a study of which professionals outperform their peers. My argument is modest: when a tool makes a plausible first draft easier to obtain, it is worth spending more attention on the question that draft is answering.
AI can lower the effort of drafting, summarising and exploring possible answers. It does not remove the work of checking facts, understanding context or deciding what matters. A fluent answer can still respond to the wrong problem.
Three questions to ask of an AI answer
- What is this output assuming I already believe?
- What has it optimised for that I did not ask it to?
- What would need to be true for this recommendation to be wrong?
These questions are useful because they make assumptions available for discussion. They do not guarantee a correct answer, and they do not require rejecting the tool. They make it easier to use the output as a starting point for judgment.
An illustrative workplace example
Imagine a team asking an AI assistant to rank two training programmes. The assistant recommends the cheaper option because it assumes cost per participant is the main goal.
The team may actually need a programme that gives beginners enough practice and feedback. The recommendation could be coherent on its own terms and still be a poor fit for that need.
A better next question would be: “What changes if successful practice by beginners matters more than the initial price?” The team can then ask what evidence supports the expected learning benefit and which information is still missing. This is a hypothetical example, not a documented comparison of those programmes.
Clarity before a better prompt
A prompt is an instruction shaped by choices: whose needs count, which constraints matter, and what an acceptable outcome would look like. Improving those choices is often more useful than asking for a more confident tone.
Plato’s account of Socrates provides a historical starting point for this interest in examining apparent certainty. Applying that approach to AI is my interpretation. It does not imply that an ancient philosopher anticipated modern models or that questioning is the only valuable skill in a changing workplace.
When an answer looks polished, try naming the decision it is meant to support. If you cannot describe that decision clearly, another paragraph of fluent output may not be the next thing you need.
Practical exercises
- Before reviewing AI output in a meeting, write down the decision you actually need to make and one assumption the output might contain.
- Compare a recommendation with the original request. Note any difference between the question asked and the problem you needed to address.
- Rewrite a recurring prompt to state the audience, the constraints and the evidence you need. Compare the two outputs, checking their claims independently.
One thought to take away
Treat an AI answer as something to examine, rather than something its fluency has already proved.