Here's a little mind reading trick. Try answering the following questions as quickly as possible and then scroll down to compare your answers with Chat GPT's.
What is the square root of 9?
Think of a vegetable.
Think of a fruit.
Think of a colour.
Think of a tool.
Scroll down
↓
Chat GPT's answers
3
Carrot
Apple
Blue
Hammer
With the exception of the colour (I was expecting Red) these are the same answers that most people give. Why is this? The usual explanation is that answering quickly and following some intellectual effort to think of these root of 9, most people pick a sort of default representative of the category asked for (the fact that a Carrot is a 'root' vegetable gives it a bit of a start too). The trick, often called the Carrot Trick, is very common (29 million Google hits), and numerous variants exist.
So then I asked Gemini AI and it gave all the expected answers. But unfortunately I made the mistake of sending all the questions as a block, so it was able to locate a Reddit post about the Carrot Trick, and just reproduced the answers it found there. But Hammer was not mentioned in the Reddit discussion, so Gemini must have come up with that independently.
There is an interesting discussion of the trick on BBC Future and there it's suggested that the trich tells us something about how our minds work; under time pressure when asked to pick a vegetable we just pick the one that stands out as the archetypal example.
That can't quite explain why Chat GPT gave most of the same responses though. I suppose must be because it has formed a strong association between 'vegetable' and 'carrot'. So is it working differently here from us? I would argue not.
It looks like there is also a cultural dimension. When I asked the Chinese AI DeepSeek the same questions it answered Broccoli, Mango, and Teal, although it still got Hammer.