Skip to main content
All posts

Does AI really think, or does it just pretend?

An AI chat sometimes seems to really think things through. What happens when a chatbot reasons step by step, and why is that different from human thinking?

You ask a chatbot a tricky question and within seconds you get a clear answer, sometimes even with a neat step-by-step plan. It quickly feels as if the AI has really thought it through. You may recognise this: you ask for help with a difficult email, a schedule for a busy week or a recipe for the leftovers in your fridge. The chatbot seems to understand your question, weigh the options and give you fitting advice. But does AI really think, the way you do?

The short answer: AI can imitate a form of reasoning and tackle problems step by step, but that is something different from human thinking. People think from experiences, memories, feelings, goals and consciousness. An AI chat isn't hungry when it suggests a recipe, isn't stressed by a full calendar and doesn't feel proud when an answer works out well. At the same time, it's interesting to explore why AI sometimes comes across as so convincingly smart.

What are the thinking messages in an AI chat?

Sometimes an AI chat first shows a plan, a short explanation of its intermediate steps or a message like: “I'll compare the options first.” These texts are often called thinking messages. They give you a glimpse of how the chatbot builds its answer, or at least of how it explains that process to you.

That's useful. Say you ask which laptop is best for working from home. A chatbot can first list what matters: your budget, how many programs you use at the same time, whether you're often on the move and how important a long battery life is to you. Then it can give its advice. Those intermediate steps show you why the answer looks the way it does. It also makes it easier to say: “Wait, I actually want a bigger screen” or “I mostly use simple programs.”

Still, a visible thinking message is not the same as a literal recording of everything that happens inside the system. It is mostly text meant to make the answer understandable. A chatbot can show a plan because that's helpful for you, just as a colleague might first sum up out loud what they're going to do. But that doesn't mean you see every small process that plays a part in producing the answer.

That's why it helps to see thinking messages as an explanation, not as proof that AI has an inner world of thoughts. They can be useful, clear and even impressive. But they don't tell you that the AI doubts, feels something or experiences an idea the way you do.

So how does an AI chat produce text?

A modern AI chat looks at your question, at what was said earlier in the conversation and at examples and patterns from the enormous amount of text it learned from. It then keeps predicting which next piece of text fits best. That's how it builds, step by step, an answer that matches your question.

That may sound simple, but it can produce surprisingly powerful results. If you ask for a friendly reminder to a client, the AI recognises, for example, that you probably want a polite tone, that you want to mention a clear agreement and that the text shouldn't be too long. It doesn't need to know what it's like to wait for a payment in order to write a useful email.

That is exactly where an important difference with human thinking lies. The chatbot works with language and with the patterns it has learned. It can sound very much as if it understands something, because language is full of clues about how people reason, explain and discuss. But being able to talk well about a subject is not automatically the same as experiencing it or truly understanding it the way a human does.

Why working step by step often helps

If you break a big question into smaller parts, you're often more likely to get a useful answer. That goes for people and for AI. Think of planning a team outing. If you just ask “Come up with a fun outing”, you might get a generic list. If you first ask about the number of colleagues, the budget, the travel time and whether everyone needs to be able to join in, the suggestion becomes a lot more practical.

A chatbot can suggest those intermediate steps itself. It might ask: “How many people are you?”, “Are you looking for something indoors or outdoors?” and “Do you want to be active or have a relaxed dinner together?” By gathering more context first, the AI has to guess less. The final answer then fits better with what you actually need.

Take a recipe. You ask: “What can I cook tonight?” A less useful answer is a random dish that calls for fancy ingredients you don't have. A smart chatbot can first ask what's in your fridge, how much time you have, whether you want a vegetarian meal and what you feel like eating. Maybe you tell it you have peppers, rice, chickpeas and yoghurt at home and fancy something spicy, but not too hot. Then the chatbot can make a fitting suggestion, including steps and substitutions.

Those questions are not a sign that the chatbot is hungry or curious the way a person is. They are, however, a practical way to get from a vague question to a more specific answer. Working step by step helps the AI take relevant information into account, organise different options and avoid falling back on a generic answer.

A good plan can still be wrong

Visible reasoning has another limitation: a logical-sounding plan isn't always correct. AI can build an answer neatly and still make a wrong assumption. For example because your question was ambiguous, because important information was missing or because the chatbot made a wrong connection.

Say you ask how long it will take to get to an appointment. The AI can show a tidy calculation, but if it assumes the wrong departure time or doesn't account for roadworks, the outcome is of little use. A neat explanation doesn't make an answer reliable by itself. So check important information, especially on topics like money, health, appointments or legal decisions.

That doesn't mean you should avoid AI. On the contrary, you can put AI to smart use as an assistant that thinks along with you. Let it write a first draft, suggest alternatives, improve a text or come up with questions you hadn't thought of yourself. See the answer as a starting point to build on, not as the final word.

So does AI really think?

That also depends on what exactly you mean by thinking. If thinking means being able to break a problem down, combine information and arrive at an answer, then AI can do something that looks a lot like it. A chatbot can imitate reasoning and help you with complex tasks.

If thinking also involves your own experience of the world, feelings, intentions and consciousness, then AI is something different from a human. A chatbot can write a comforting sentence without ever knowing grief. It can compare routes without ever having travelled anywhere. It can explain how to have a good conversation without ever experiencing one.

That keeps the philosophical question interesting. Perhaps our definition of thinking will change as AI gets better. For now, it's mostly useful to remember that a convincing answer and genuine human consciousness are not the same thing.

Try it yourself

Want to see what working step by step does? Ask a chatbot a practical question from your own work or daily life, for example: “How can I organise my inbox better?” First ask it directly. Then ask the same question again, but add: “Think step by step first and ask me questions if you're missing information.” Compare the answers. You'll probably notice that the second answer fits better, precisely because the AI first gets room to explore the question.

Use that result as an invitation to keep experimenting, stay critical and discover where AI can help you in practical and creative ways. When does an answer from AI feel like real thinking to you, and when do you clearly notice that it's something else?

Have a project in mind?

I'd love to hear about it. Send me a message and I'll get back to you as soon as possible.

Get in touch