“why has the man in my ai picture got three arms 😭”

This is a really common one, and once you know why it happens, you’ll start spotting it everywhere.

When you ask an AI to make an image, it isn’t drawing in the way a person draws. It doesn’t know what a person actually is, or that people have two arms, or that hands have five fingers.

Here’s what it’s doing instead. A digital image, zoomed right in, is made up of millions of tiny dots of colour. What the AI has learnt, from the millions of pictures it was trained on, is the patterns in those dots. Which colours tend to sit next to each other to make a face. What pattern of dots tends to form a kitchen. It’s generating dots that fit a pattern, not drawing a scene the way you or I would.

So it’s learnt roughly what “a man standing in a kitchen” tends to look like. It’s picked up the general shape, the lighting, the overall look. But it doesn’t have a rule that says stop at two arms. It just keeps generating whatever looks plausible, and sometimes that includes a third arm, six fingers, or a door that opens into a wall.

It’s getting better quickly. But this is why it still goes wrong sometimes. Not because it made a mistake, but because it never actually knew what it was drawing in the first place.