It is remarkable that over a quarter of humans didn't successfully identity other humans!
A forced-choice test where an Interrogator interacts with an AI and a human and picks which one is more likely to be human would probably (hopefully!) have a much higher accuracy rate.