That works if the problem is a self-selection bias. If there are counterproductive selection biases whose source isn't the selectees, or even if there's just an insufficient net bias in the right direction (what tiny fraction of people wouldn't be nearly useless at solving global health problems?) it's going to be harder to solve.