The argument was in response to the implicit assumption that even if the tests are biased, we shouldn't worry about that as long as they can predict performance. "Imagine" takes the place of the word "if" in there and the intended conclusion is "yes, you should worry about whether the tests are biased, because if they are biased, that would be bad." To make that conclusion I do not need to provide evidence that the imaginary scenario applies to reality.

If I say "imagine that you jumped off a cliff. You'll get smashed. Maybe you should avoid jumping off cliffs", I don't need to provide evidence that you jump off cliffs, because the conclusion is in the form of a conditional that already conditions on whether the imaginary scenario is real.