Point taken, but to be clear: bias doesn't imply racism, no matter how hard you might want it to.

Like airplane accidents, algorithms accidents are products of whole organizations, which include engineers, but are not only engineers. I don't condone blaming individual engineers for incidents when there is an entire system of quality control and management surrounding them that is supposed to proactively prevent these kinds of things. Nonetheless, it is possible for the whole system to be negligent, and it's also possible for the whole system to, if not be racist (because organizations don't have sentiments), to act collectively in a way that aligns with what a racist would do, if they were an organization. I admit that it's a stretched analogy and borders on anthropomorphizing organizations, but at the same time it could be a useful conceptual tool.

For a concrete example, imagine that a company was prioritizing investigation of moderator actions based on the average emotional reaction of twelve or so people on an accountability board, when presented with each case. If the accountability board has any biases that don't average to zero across the dozen, the diverse stream of injustices will get corrected more in one direction than another, and the result will be injustice even in the absence of any individual with bad intentions.