True. When displaying an image your PC doesn't "read" it, as in interpret the contents of the image.
But algorithms can be created that does that.
It's all about understanding the information in relation to its surroundings.
<Blue pixel here, green pixel there. 65% chance its a leaf against a sky background (simplyfied but the gist of it)
As i said. I've written bots specifícally to solve captchas so I know it is possible.
With captchas like the ones here it's especially easy because you can request a new captcha at any time and it's only letters so it just have to approximate shapes limited to 26 possible shapes.