Sometimes I’m chatting with ChatGPT and catch myself thinking:
“Did you actually understand that?” 😄 Because it doesn’t just answer.
It connects ideas, refers back to things I said earlier, and sometimes notices a link I hadn’t even thought of myself.
In moments like that, it is surprisingly easy to forget that there isn’t another person sitting on the other side of the conversation.
But if AI doesn’t think like a human, what is it actually doing?
And can we even call what it does thinking?
What happens when we think?
That question is not especially simple either.
If I’m deciding what to cook for dinner, for example, I’m not relying on one piece of information.
I remember what we have at home. I know what we like. I remember what we ate yesterday. I look in the fridge and notice what needs using up soon. I might suddenly remember something my mum used to cook twenty years ago.
And somehow, from all of that, a decision appears.
Human thinking is shaped by our experiences, our senses, our memories, our bodies, our emotions, our goals and everything we have learned about the world throughout our lives.
Modern artificial intelligence works very differently.
AI learns patterns
Put very simply, a modern language model learns from enormous numbers of examples and gradually builds up relationships between words, concepts, sentences and ideas. It doesn’t sit down with a book, read it and then say:
“Right. I understand. I’ll remember that.”
Instead, it learns huge numbers of patterns and connections. Which words tend to appear together? Which concepts are related? What kinds of conclusions tend to follow from certain information? What does a good explanation, argument, summary or even a joke look like?
Then, when we ask it something, it works with those learned relationships. That is why it can generate a new answer rather than simply retrieving a sentence it has stored somewhere.
So does it just predict the next word?
You hear this explanation quite often, and there is truth in it.
One of the foundations of language models is indeed predicting what should come next.
But if we stop there, it is a little like describing human speech by saying:
“We just move our mouths.” Technically true. But it leaves out rather a lot. 😄
During training, modern models develop extremely complex relationships between concepts. That allows them to work through problems in several steps, recognise patterns, compare possibilities and generate new responses.
That is far more interesting than simply “guessing the next word”.
But it still doesn’t mean they think the way we do.
So what is the biggest difference?
AI doesn’t have a life story like you or I do. It didn’t have a childhood. It has never burnt its hand on a hot baking tray and become more careful the next time. 😄 It has never stood in a bank queue. It has never fallen out with a colleague. It has never sat opposite a customer and had that feeling that something about the conversation just isn’t quite right.
We understand a huge number of things because we have lived them.
AI, on the other hand, works with what it learned during training and the information available to it in the current conversation or task.
That is also one reason it can sometimes be very confidently wrong. A human may stop because something simply feels off. We should not automatically assume that an AI system has that same kind of human common sense.
But does it understand what we say?
This is where the answer becomes more interesting than a simple yes or no.
AI can handle meaning and relationships between concepts in very sophisticated ways.
If I tell it: “My customer is angry because they have had to explain the same thing three times.” it can recognise why that customer is frustrated and suggest an appropriate response.
In one sense, that can look very much like understanding. But it is not the same kind of understanding I have when I recognise that frustration because I have been a frustrated customer myself.
And that difference matters.
Why does this matter in business?
Because it is very easy to fall into one of two extremes.
The first is: “It’s just a stupid computer programme.”
It isn’t. AI can be remarkably useful for the right kinds of tasks.
The other is: “It’s intelligent. It’ll work out what to do.”
That is where I would be much more careful.
If we use an AI system for a real business task, we should not assume that it will apply judgement in the same way an experienced human colleague would.
We still need to define what it knows, what information it can use, what decisions it is allowed to make, what it must not do, and when a human needs to step in.
Precisely because AI can sound extremely intelligent without thinking like a human being.
And that leads to another question that I find just as interesting.
If there isn’t actually a person on the other side, if it has no personal life experience, and if it doesn’t think the way we do… who am I actually talking to when I open ChatGPT?
What is happening in the background when I type a question and, a few seconds later, a completely natural-sounding answer appears?
And who exactly is the “I” when ChatGPT says: “I think…”? 😄
That is what we’ll look at next week.