Should AI Give Us Answers—or Help Us Make Better Decisions?

 

Many people tell me that the biggest problems with generative AI are that it is undermining our ability to think for ourselves and that we cannot trust its output.

These are legitimate concerns. If we increasingly rely on AI to do our thinking for us, do we risk losing the ability to question, analyze, and make decisions independently? And if AI can produce convincing but incorrect answers, how can we know when to trust it?

But I started wondering: what if we could create a situation where the opposite is true? Could the mistakes made by AI become an opportunity to strengthen our critical thinking rather than diminish it?

Perhaps the answer lies not just in improving AI, but in changing how we use it.

What are hallucinations?

I love the term hallucination. The idea of seeing something that isn't really there seems to capture perfectly what happens when a generative AI system comes up with a convincing but ultimately incorrect answer.

AI can produce information that sounds authoritative, even when it has no reliable basis in fact. This means we need to be skeptical of its output, particularly when accuracy matters.

But I wonder whether there is another way to look at these mistakes.

What if we sometimes treated an unexpected AI response as an invitation to explore a different possibility? Could an incorrect answer prompt us to consider an alternative explanation, challenge an assumption, or imagine a different version of reality?

Of course, we must not confuse an invented answer with an alternative truth. In situations where factual accuracy matters, hallucinations can be harmful. But in creative exploration, an unexpected response might occasionally spark an idea that would never otherwise have occurred to us.

The opportunity, then, is not to celebrate mistakes for their own sake, but to learn how to use unexpected outputs without being fooled by them.

How can we reduce the risks of hallucinations?

One approach is to stop asking AI for a single answer and instead ask it to generate several possibilities and compare them.

For example, rather than simply asking how to reduce AI hallucinations, we could use the following prompt:

Prompt:

Compare several methods for mitigating AI hallucinations. Explain how each method works, its strengths and weaknesses, and how effective it is. Include approaches that encourage creativity and active human decision-making. Identify the best overall approach and explain why, highlighting any limitations or trade-offs.

This changes the nature of the interaction. Instead of receiving one answer and deciding whether to accept it, we receive several perspectives that we can examine, question, and compare.

We can look for differences, identify assumptions, weigh the advantages and disadvantages, and decide which approach makes the most sense.

Of course, asking for multiple answers does not automatically make them more accurate. AI might repeat the same mistake in different ways, or present several equally plausible but incorrect explanations. The value comes from comparing the answers critically, rather than assuming that the majority must be right.

We can strengthen this approach in other ways, too. We can provide AI with structured, reliable sources of information, ask it to cite its evidence, and check those citations ourselves. We can also ask it to critique its own answers, identify what might be wrong, and explain where uncertainty remains.

Each technique has limitations, but together they can help us move from simply accepting an answer to actively evaluating it.

An example: Asking AI for an answer versus asking it to help us think

Imagine you run a small business and want to attract more customers.

Approach 1: Ask AI for an answer

You ask:

“What is the best way to attract more customers to my business?”

AI might recommend investing in social media advertising, improving your website, or offering discounts.

The answer sounds reasonable, and you might be tempted to follow its advice. But how does AI know which approach is best for your business? It may know nothing about your customers, your budget, or your competitors.

A convincing answer is not necessarily the right answer.

Approach 2: Ask AI for alternatives and comparisons

Now imagine you change your prompt:

Prompt:

Suggest five different ways a small business could attract more customers. For each approach, explain its strengths, weaknesses, costs, assumptions, and potential risks. Identify what additional information would help determine which approach is best. Then critique your recommendations and explain what could make each one fail.

The interaction becomes much more interesting.

AI might suggest social media marketing, referral schemes, local partnerships, search engine optimization, and introductory offers. Rather than simply recommending one, it compares them and highlights the circumstances in which each might work.

You now have something to think about.

Perhaps social media advertising offers a wide reach but requires a budget and ongoing effort. A referral scheme might cost less but depend on having satisfied existing customers. Local partnerships might be particularly effective if your business serves a specific geographical area.

You can question the assumptions, ask for evidence, and consider which options suit your circumstances.

Approach 3: Make the decision together

Finally, you ask:

Prompt:

Based on these alternatives, what information would you need to recommend an approach? Identify the biggest uncertainties, suggest a low-cost way to test the most promising options, and explain what results would indicate that an approach is working.

Now AI is helping you develop a decision-making process rather than simply supplying a recommendation.

You might decide to test a referral scheme alongside a small social media campaign, measure the results, and compare the cost of acquiring each new customer before investing more money.

The final decision is yours, informed by evidence rather than by AI's confidence.

What has changed?

The AI system has not necessarily become more intelligent or more accurate. What has changed is the way you are using it.

You have moved from accepting a recommendation to exploring alternatives, questioning assumptions, and testing ideas.

And there is a further benefit: even if AI makes a mistake or overlooks an important factor, you have more opportunities to notice it because you are actively examining its output.

We are no longer simply asking AI to think for us. We are using it to give us more things to think about.

Is generating multiple answers the best approach?

In my opinion, generating multiple answers is particularly promising because it can support two things that are often overlooked in discussions about AI: creativity and human decision-making.

First, it encourages creativity. By asking for different approaches, we create space for alternative perspectives and unexpected connections. We might reject most of the suggestions, but one could trigger an idea that is genuinely useful.

Second, it encourages us to become active participants in the decision-making process.

When AI gives us a single, confident answer, it is tempting to accept it, especially if it sounds convincing. When it presents several alternatives, we have to think about what distinguishes them. Which assumptions are reasonable? What evidence supports each option? What are the trade-offs? What matters most to us?

The decision becomes something we participate in, rather than something we simply receive.

I would take this a step further by asking AI to critique its own answers and explicitly identify uncertainty. Rather than asking only which option is best, we could ask what might be wrong with each option, what evidence is missing, and what would change the recommendation.

This is not a guarantee of accuracy. AI can fail to identify its own errors, and its confidence is not proof that it is right. But the process gives us more opportunities to question the output and check the claims that matter.

From answer machines to thinking partners

Perhaps we should stop thinking of AI primarily as a machine that provides answers and start thinking of it as a tool that helps us explore possibilities and make better decisions.

That would mean using AI differently depending on the task. If we are brainstorming, we might welcome unusual suggestions and explore several possible directions. If we are making a decision that depends on facts, we would put greater emphasis on reliable sources, independent verification, and evidence.

In both cases, our role remains important. AI can generate possibilities, highlight trade-offs, and challenge our assumptions, but we still need to decide what to believe and what to do.

There is a wider question here, too. If we design our interactions with AI to encourage questioning, comparison, and reflection, could we make people more engaged in their own thinking rather than less?

That is something worth exploring. It would be premature to assume that these techniques automatically improve critical thinking over the long term, but they offer a practical way to make our interactions with AI more thoughtful.

A better way to use AI?

I started with the concern that AI might undermine our ability to think for ourselves. I wonder whether the answer is not simply to make AI more accurate, but also to change the way we ask it questions.

Instead of asking for the answer, perhaps we should ask for several possible answers, compare their strengths and weaknesses, examine the evidence, and make our own decisions.

The goal is not to trust AI blindly, nor to reject it because it sometimes makes mistakes. It is to use it in a way that encourages curiosity, creativity, and critical judgment.

Perhaps the real opportunity is not to make AI do our thinking for us, but to use it to help us think better.

So, should AI give us answers—or help us make better decisions?

Acknowledgement

This blog reflects my own thoughts, opinions and analysis. AI was used solely to improve grammar, readability and structure while preserving my original ideas and writing style.