AI Isn't Useless—You Just Don't Know How to Use It

Recently, I read an article saying that the adoption rate of generative AI is lower in Japan than in other countries.

The reasons it cited included “AI is difficult to use,” “it sometimes lies,” and “it does not give me the answer I expect.”

AI certainly does lie. It sometimes presents nonexistent information as if it were real, and it can misunderstand our intentions and return an outrageous answer.

But as someone who uses AI every day, I also find myself thinking:

Perhaps it is not that “AI is useless.” Perhaps people simply do not yet know how to use it.

I often see people online saying, “AI does nothing but lie, so it is useless,” or “It only gives vague, unreliable answers.”

But AI is neither a search engine that always returns the correct answer when you enter a question nor a mind reader that understands your intentions without any explanation.

It is a tool that also requires a certain degree of skill from the person using it.

AI Is a Mirror of Its Input

The answers from generative AI change dramatically depending on the information and instructions we provide.

“Recommend some books.”

Ask only that, and the AI has little choice but to make a generic selection from books that are widely praised.

But ask:

“Recommend five books for an IT engineer in their fifties who wants to think about their career in the age of AI. Do not choose technical manuals; choose books about changes in work and society.”

The answer will be quite different.

The more clearly we communicate the purpose, background, constraints, and desired format, the more accurate the answer becomes.

The IT world has long had a saying: “Garbage In, Garbage Out.” Put garbage in, and garbage comes out.

Few tools demonstrate that principle as clearly as generative AI.

“AI Lies” Is True—That Is Why You Need to Know How to Use It

Of course, hallucinations are a real problem with generative AI.

But it is a leap to go from “it sometimes lies” to “it is useless.”

I use AI to write prose and programs, and I also use it for research.

But if a fact is important, I check the primary source. If it writes a program, I test it. If an answer looks suspicious, I ask for the evidence. If necessary, I ask another AI the same question.

I never treat AI as a machine that is absolutely incapable of making mistakes.

We do not unconditionally trust a document prepared by a human subordinate or colleague either. If a number looks wrong, we check it. If the decision is important, we consult the original source.

For some reason, however, people suddenly demand a perfect score from AI: “If it lies even once, it is useless.”

Apply that standard to humans, and there would be nobody left in the company.

Vague Instructions Can Produce Only Vague Answers

Giving instructions to AI is much like assigning work in the office.

Imagine a manager who tells a subordinate:

“Just put together some kind of nice presentation.”

The next day, the manager looks at the result and becomes angry:

“What is this? It is completely useless.”

What would you think of that manager?

The manager explained neither the purpose, the intended audience, the number of pages, nor the message the presentation was supposed to convey.

If the result failed to match expectations, was it really entirely the subordinate’s fault?

The same thing happens with AI.

Provide the background. Explain the purpose. Set the constraints. If the answer is wrong, tell it what to correct.

Instead of trying to obtain a perfect answer in one attempt, use dialogue to guide it toward the destination.

Once you understand this, AI becomes a remarkably useful tool.

You Do Not Need to Memorize “Magic Prompts” Either

That does not mean you need to memorize a collection of stock phrases such as, “You are an excellent expert.”

It is much simpler than that.

Explain what you want to accomplish in terms the AI can understand.

If information is missing, add it.

If the answer is wrong, say, “That part is incorrect.”

If it is too long, tell it to “cut it by 30 percent.” If it is too abstract, tell it to “add a concrete example.”

The ability to use AI well is not the ability to memorize special incantations.

It is closer to the ability to put the purpose and constraints in your own mind into words and communicate them to someone else.

The Capability of a Tool and the Ability to Use It Are Different Questions

AI has flaws.

It lies. It misunderstands. It does things nobody asked it to do.

That is true.

But if we conclude from this that it is therefore useless, the discussion ends there.

Calculators, spreadsheets, and search engines all became tools of work in the hands of people who learned how to use them. Generative AI alone does not need to be something everyone can use perfectly from the beginning without practice.

If anything, I think AI is a tool that reflects the ability of its user with unusual clarity.

Vague questions produce vague answers. Specific questions produce specific answers.

So the next time you feel like saying, “AI is useless,” it may be worth asking yourself one question.

Is AI truly unusable, or have you simply not learned how to use it yet?

If one reason for Japan’s low AI adoption rate is that “AI is difficult to use,” there is a certain irony in that.

While some people wait for AI to become easier to use, the people already using it continue to get better at it.