Does it count if you allow the test to be accurate but you still maintain control over when it is wrong?

I just spammed F and randomly hit D and it pretty much never predicted the D.

Which I think captures a fundamental flaw in algorithms vs humans. Humans know how to mask their signals within noise. I'm inclined to believe the generalization that algorithms that aim for accuracy are highly susceptible to being manipulated by humans who are allowed to have free will over the inputs.