From Better Words to Better Judgment

Once your thinking is visible on the page, another question comes into focus: among the answers an AI can produce, which one is actually good?

Imagine a neighborhood bookstore asking an AI to write a short announcement for a children’s reading night. It offers two drafts. The first is energetic and glossy: “Unlock a world of wonder! Join us for an unforgettable literary adventure that will spark imagination and inspire young minds.” The second is quieter: “Bring a book you love—or come choose one here. On Thursday, we’ll read together, make room for questions, and send every child home with a story.”

Both are fluent. The first has polish, rhythm, and the reassuring air of something written by a marketing department. The second is more specific. It tells families what will happen and gently removes a worry: a child need not bring a book. For this bookstore, with its welcoming atmosphere and many first-time visitors, that small reassurance may matter more than the grand promise of an “unforgettable literary adventure.”

Noticing that difference is taste.

Taste is often mistaken for personal preference: I like this; I don’t like that. Preference matters, but useful discernment goes further. It asks whether a piece of work suits its purpose, its audience, and its setting. A playful headline might be right for a children’s event and wrong for a notice about a serious safety issue. A concise answer may be elegant—or may leave out the one detail a reader needs.

AI can make an option sound finished before it is well judged. Fluency is easy to admire because it is immediately visible. Relevance is quieter. It may depend on knowing that the bookstore wants new families to feel welcome, that the event is free, or that parents need to know whether they can stay. If you have never thought about the audience, a smooth draft can pass inspection simply because it reads smoothly.

Experience helps you notice what a newcomer might miss. A teacher may recognize that “engaging activities” tells parents almost nothing; a specific description of the activity gives them a reason to come. A shop owner may see that a clever slogan obscures the actual offer. Experience is not infallibility. It is a store of comparisons—examples of what worked, what confused people, and what sounded impressive but fell flat.

Clear goals sharpen that judgment. Before choosing between drafts, ask: What should the reader understand, feel, or do? Who might feel excluded or uncertain? Which detail would make this more useful? These questions turn a vague sense of “better” into criteria you can explain. They also give the AI something to work with: not merely “make it more engaging,” but “make first-time families feel welcome, and state clearly that children may borrow a book at the event.”

You do not need expert taste in every field to begin. You need attention, some knowledge of the purpose, and a willingness to compare. Ask for alternatives, then look past the shine: Which version is more accurate? Which speaks to the real audience? What is missing? What sounds generic because it could belong to anyone? That is the work of inspecting an answer for relevance, omissions, uncertainty, and fit—not just polish.

Taste develops as you acquire examples and learn to name distinctions. The point is not to become a harsher critic of machine-made prose. It is to become a more attentive reader of it—and to recognize that a polished answer is an invitation to inspect, not a reason to stop.

Once you have inspected an answer, the next step is deciding what, if anything, to do with it.

Imagine the AI has helped a neighborhood library draft a plan for attracting families to a weekend event. The suggestions sound sensible: offer refreshments, add a craft table, promote the event on social media. But perhaps the library has no budget for food, the room is too small for messy activities, and the families it most wants to reach rarely see its social media posts. The plan may be fluent and relevant, yet wrong for this library.

That gap matters because an AI suggestion is not yet a decision. It is an input—a candidate to weigh against evidence, people, and circumstances the system may not know.

Check what supports the answer

Begin by separating what the answer claims from what would make the claim trustworthy. If it recommends a tactic, ask: Is this based on information we provided, a source we can inspect, or a general pattern that may not fit here? For factual claims, look for evidence and verify consequential details against reliable sources. A confident tone is not a source. Nor does a specific-sounding explanation prove that the model has checked anything.

Sometimes the useful follow-up is simple: “Which parts of this proposal depend on assumptions? List them separately.” That can expose hidden premises—perhaps that parents have time on Saturday, that a particular channel reaches them, or that a free event needs no explanation of its value. The AI can help surface assumptions; people still need to test them.

This is not a demand to distrust every sentence. It is a way to match scrutiny to stakes. A suggested headline needs a lighter check than advice affecting someone’s health, finances, rights, or safety. The more consequential the choice, the less acceptable it is to treat a plausible answer as sufficient evidence.

Ask who bears the consequences

A decision is not good merely because it solves the problem as phrased. It also matters who benefits, who carries the cost, and whose needs were left out of the framing. A proposed school policy might save staff time while making participation harder for students with limited internet access. A hiring shortcut might appear efficient while screening out qualified applicants for reasons unrelated to the job.

AI can help widen the view: “Who could be disadvantaged by this option?” or “What might we be overlooking about the people affected?” These are prompts for thought, not guarantees that every perspective will appear. The people involved, and those who understand their circumstances, remain essential sources of context.

In the library example, families might value a quiet, welcoming space more than a craft table. A staff member who knows the neighborhood may recognize that a partner organization—not another online post—is the better route to reaching them. Context is not decorative background. It can change what counts as a good answer.

Choose, adapt, or decline

After checking the evidence and widening the frame, the human task is to choose an action proportionate to the situation. Sometimes the answer is ready to use. Often, it needs adapting: keep the clear invitation, replace the costly refreshments with water, and ask a community partner to share the event. Sometimes the responsible choice is not to proceed until a missing fact is checked.

A useful final question is: Would I be willing to explain why we chose this, to the people affected by it? If not, the reasoning may be incomplete—or the choice may need to change.

The AI can generate options quickly, compare trade-offs, and help articulate a rationale. It cannot take responsibility on behalf of the people who must live with the result. Its answer may be intelligent in form and still lack the local knowledge, values, or accountability that make a decision fit for a real situation.

Keep the distinction clear: a model can suggest; a person or institution must decide. Judgment is not a ceremonial human signature added at the end. It is the work of connecting an answer to evidence, consequences, and the particular lives in front of us.

Return to the two colleagues preparing the project update. Both can now produce a professional paragraph in seconds. That achievement no longer tells us much about the quality of their work. The important difference is whether the paragraph preserves what the team knows, reveals what it does not know, and helps someone make the next decision.

The colleague who noticed the provisional delivery date did something small and consequential. They read a word carefully, understood the distinction it carried, and wrote an instruction that kept the distinction alive. When the answer arrived, they had a standard against which to judge it. Reading and writing had shaped the work from beginning to end.

These abilities grow through use. The next report can teach you a concept you did not have. The next difficult paragraph can reveal where your understanding runs out. The next brief can force you to decide whether you want a summary, an explanation, or a recommendation—and what evidence each would require. Every such encounter gives you something to bring to the machine that was not there before.

Reading more is valuable when more of what you read becomes available to your thinking. A term you can apply, an argument you can question, an example you can compare with a new situation: these are useful gains. A pile of completed articles is a poor substitute for one distinction you can now see clearly.

Writing better means making those distinctions visible to someone else. It can be as modest as changing “the project is on track” to “the internal work is complete; the supplier's delivery date remains unconfirmed.” The second sentence is less comfortable. It is also more useful. It tells the reader where certainty ends and what deserves attention next.

AI can help you practise both skills. It can offer an explanation, challenge a paraphrase, suggest a different structure, or expose an ambiguity. But let it give you something to work on. Read the passage yourself. Try your explanation before asking for one. Decide which revision preserves the meaning and be able to say why. The work you do in that exchange is part of what you take into the next one.

You do not need to turn every conversation into an examination. Explore freely when the stakes are low. When an answer will guide a consequential choice, connect it to evidence, circumstances, and the people affected. A fluent response remains a contribution to judgment; you must decide what it is worth.

Start with the next real task. Read one relevant passage closely enough to explain it in your own words. Write what you need clearly enough that another person could recognize a useful answer. Then inspect what comes back. Notice what this reveals about the task—and about your own understanding.

For decades, becoming more capable with computers meant learning more about how machines worked.

That knowledge still matters. But something remarkable is happening: computers are increasingly capable of working with the language we already speak.

And this makes some of our oldest intellectual skills newly powerful.

Every concept you learn gives you another way to describe a problem. Every difficult argument you learn to understand gives you another perspective to bring to a decision. Every sentence you learn to write clearly makes your intentions easier to communicate—to another person or to an intelligent machine.

The next generation of AI will probably become better at understanding our instructions and carrying them out. But it cannot decide for us what is worth understanding, which problems deserve attention, or which outcomes we should pursue.

Words are the new code. Understanding is what makes them powerful.

Read more. Think deeper. Write better.