This begs the question, how do you define a bot? From my understanding of the talk, they replay one person's session and don't use composites. But, point taken.
For the ESP Game, it might be the case that they replay one person's session.
But for Phetch:
"Emulating Seekers in a convincing manner is more difficult. If a real player enters useless
descriptions of the image, we do not want the emulated Seekers to find it. Although this is not a significant
problem since most Describers enter accurate descriptions of images, we nevertheless address it to protect
the illusion that a real game is being played. The solution to the problem relies on the fact that we are using
images from the ESP Game database, so the keywords associated with the images in the search engine are
already known. Therefore, we emulate Seekers by not guessing the right image until a certain number of
the ESP keywords have been entered by the Describer."
But for Phetch:
"Emulating Seekers in a convincing manner is more difficult. If a real player enters useless descriptions of the image, we do not want the emulated Seekers to find it. Although this is not a significant problem since most Describers enter accurate descriptions of images, we nevertheless address it to protect the illusion that a real game is being played. The solution to the problem relies on the fact that we are using images from the ESP Game database, so the keywords associated with the images in the search engine are already known. Therefore, we emulate Seekers by not guessing the right image until a certain number of the ESP keywords have been entered by the Describer."
http://reports-archive.adm.cs.cmu.edu/anon/2005/CMU-CS-05-193.pdf#page=56