It's really not; disparate impact deals with it, but the doctrine was articulated in a case with ample evidence that the policy at issue, though facially neutral, was adopted as a deliberate replacement for explicit racial discrimination with an intent to maintain it's discriminatory effect and without other business purpose, and the doctrine was clearly intended principally to address such veiled-substitute policies.
The Supreme Court did not hold that, it merely noted that the lower courts had held that, and noted that a lack of discriminatory intent was “suggested” by the Company's efforts to help undereducated employees by financing a major portion of remedial education (part, but not all, of the policy held to be disparate impact was adoption of high school diploma requirements for positions for which such requirements has no substantial relation to job performance.)
The Supreme Court explicitly did not consider whether there was intent to discriminate, because it was not necessary to do so to resolve the case.
Not if what the algorithm optimizes is a measure provably closely linked to job performance.
In fact, as the array of protected classes and the statistical means available will probably over time allow proving prohibited disparate impact for any hiring method other than an algorithmic method optimizing a true job performance measure, algorithmic methods may soon be practically mandatory.
OTOH, algorithmic methods targeting any figure of merit that isn't provably tightly linked to job performance are going to be problematic, sure.