Treat a chatbot like you'd treat a person—skeptical, curious, willing to ask again.
As artificial intelligence becomes woven into the fabric of daily decision-making, educator Tom Hewitson offers a corrective to the two most common errors humans make with tools like ChatGPT: uncritical faith and premature abandonment. His insight is ancient in spirit — that wisdom lies not in the tool itself, but in the quality of the relationship we form with it. To engage well with AI, as with any imperfect source of knowledge, is to bring the same skeptical curiosity we owe to all human-generated information.
- Millions of people are either walking away from AI after a single bad answer or accepting its outputs as unquestionable truth — and both habits are quietly causing harm.
- The core tension is a misunderstanding baked into how we approach these tools: we treat them as oracles rather than as fallible synthesizers of existing, sometimes flawed, human knowledge.
- Because chatbots learn from what humans have written — errors included — they can confidently repeat misinformation with no internal mechanism to flag the difference.
- Hewitson is pushing a different model: treat AI like a conversational partner, not a vending machine — rephrase, iterate, cross-reference, and bring your own judgment to every exchange.
- The trajectory is clear — as AI embeds itself deeper into work and learning, the gap between those who engage critically and those who don't will increasingly determine who gets misled and who gets informed.
Tom Hewitson has spent years teaching people how to use artificial intelligence, and his central finding is disarmingly simple: most of us are doing it wrong in one of two ways. We either dismiss the technology after one unsatisfying result, or we accept its answers as fact because they came from a machine. Both reactions, he argues, miss the point entirely.
The misconception runs deep. We approach chatbots like ChatGPT, Copilot, and Gemini as if they were perfect systems — infallible oracles that either work or don't. But the reality is more human than that. These tools don't generate knowledge from scratch. They synthesize information that already exists in the world, written and published by people. And if a person published something wrong, the chatbot can repeat it without knowing the difference. That's not a flaw — it's simply how the systems work.
Hewitson's practical answer is to treat a chatbot the way you'd treat a person: not as an authority to obey or distrust outright, but as a conversational partner. When someone gives you an answer that doesn't quite fit, you rephrase the question, approach it from another angle, cross-check it against what you know. The same instinct should guide AI use — ask differently, break the question apart, compare answers, look for patterns and contradictions.
The stakes are genuine. As these tools become more embedded in how we work and decide, the difference between blind faith and skeptical engagement becomes the difference between being misled and being informed. Hewitson's message is that what works is neither extreme — it's curiosity paired with verification, and iteration paired with judgment.
Tom Hewitson has spent years teaching people how to actually use artificial intelligence, and what he's learned is this: most of us are doing it wrong in one of two ways. Either we ask a chatbot a question, get an answer we don't like, and decide the whole thing is useless. Or we get an answer and treat it as gospel, assuming that because it came from an AI it must be correct. Both reactions, Hewitson argues, miss the point entirely.
Hewitson, an AI educator and expert, recently laid out the core misconception that shapes how millions of people interact with tools like ChatGPT, Copilot, Gemini, and Grok. The problem isn't the technology itself—it's how we've been taught to think about it. We approach these chatbots as if they were perfect systems, infallible oracles that either work flawlessly or don't work at all. That binary thinking is the trap.
The reality is simpler and more human than that. A chatbot doesn't generate knowledge from first principles. It synthesizes information that already exists somewhere in the world—information that humans have written, published, and uploaded. And here's the catch: if a person published something wrong, the chatbot can and will repeat it. The AI has no way to know the difference between accurate information and misinformation if both are present in its training data. It's not a flaw in the machine. It's a feature of how these systems actually work.
This is where Hewitson's core insight lands: you should treat a chatbot the way you'd treat a person. Not as an authority figure, not as a friend you can blindly trust, but as a conversational partner. When you talk to someone and they give you an answer that doesn't quite fit, you don't usually just accept it or walk away. You ask again. You rephrase the question. You approach it from a different angle. You say, "Wait, what do you mean by that?" You cross-check what they told you against what you already know or what other people have said.
That same instinct should guide how we use AI. Ask the same question in different ways. Break down what you're really trying to find out and ask it piece by piece. Have a conversation with the tool, not a transaction. Synthesize what it tells you. Compare answers. Look for patterns and contradictions. This approach transforms the chatbot from a vending machine that either delivers or disappoints into something more useful: a thinking partner that requires you to think as well.
The stakes of getting this right are real. As AI tools become more embedded in how we work, learn, and make decisions, the difference between treating them as infallible and treating them as flawed-but-useful becomes the difference between being misled and being informed. Hewitson's message is that neither extreme—blind faith or total dismissal—serves us. What works is something closer to skeptical engagement: curiosity paired with verification, iteration paired with judgment.
Notable Quotes
Many users mistakenly think ChatGPT, Copilot, and similar tools are perfect systems that solve problems flawlessly, when in fact they synthesize existing information and can repeat errors from their training data.— Tom Hewitson, AI educator
The Hearth Conversation Another angle on the story
So when you say people treat chatbots like either perfect machines or useless ones, what does that actually look like in practice?
Someone asks ChatGPT for help with a work problem, gets an answer that doesn't solve it, and concludes the whole thing is a waste of time. Or they ask for a fact, get a confident-sounding response, and never question it. Both people are missing the middle ground.
But isn't there something appealing about the idea that a machine could just be right? That's kind of the fantasy, isn't it?
Absolutely. We want certainty. But a chatbot is only as good as the information it learned from, and that information is messy and contradictory and sometimes just wrong. The machine doesn't know the difference.
So you're saying the user has to do the thinking?
Not instead of the AI—with it. You ask, you listen, you ask differently, you compare what you hear. It's more work than just accepting an answer, but it's the only way to actually use these tools well.
Does that mean people need to be trained to use AI, or is it something they should figure out naturally?
Most people aren't figuring it out naturally. They're either trusting too much or not enough. It takes a shift in mindset—treating the chatbot like a colleague who's smart but fallible, not a god or a broken tool.