I think a better strategy is to remove the actual correlation between race and academic performance, and possibly the one between race and criminality for that matter.

That is a necessary prerequisite, yes. As long as such an actual correlation is in place, it will be observed and will result in a perceived correlation.