Objections to computer-assisted proofs are remarkably similar to those expressed with regards to machine-learning models; both usually being centered on the epistemic opacity of either methodology. In this chapter, I explore how these objections differ, arguing that their differences indicate that epistemic opacity is a matter of degree. I then argue that one’s level of comfort with a given degree of epistemic opacity will be, in large part, determined by one’s position with regards to scientific realism. If one is a realist, I argue, one’s tolerance for epistemic opacity will be lower. This brings to light the bearing of the well-established debate between scientific realism and anti-realism into contemporary discussions over the acceptability of epistemic opacity in machine-learning.