What makes you think a computer run system won't be subject to biases (in particular of the implementors)?

Bureaucracy isn't the promise of a system free of biases. The promise is that of a pure function: \forall a,b: a = b -> f(a) = f(b)

What's the point? The point is that there is a clear, rational audit trail for "why was this decision taken" with a computer system. You can challenge that. You can't challenge an overpaid gov admin whose primary motivating factor in making decisions is pleasing his superiors, no matter how removed.

Highly recommend you check out the book "Weapons of Math Destruction" by Cathy O'Neil

Very accessible set of case studies where algorithmic management of traditionally human-run bureaucratic processes is biased and destructive.

There isn't any meaningful difference between the kinds of challenges you present. It's given a facade of mathematical rigor, but it's still a facade.

I agree partially - I think it's easier to audit the biases an implementor has encoded into an automated system, especially since the pressures a high-level official exerts are often verbal and unrecorded.

We can barely understand existing machine learning systems, what makes you think that we will understand those that run a global bureaucracy?

Machine learning systems by their nature design their logic without human intervention; a bureaucratic system encodes human rules in a way that a human programmer decides. cf a US college's system for signing up for classes, which is usually a one-to-one translation of preexisting bureaucratic procedure.