What does "garbage in, garbage out" mean for AI bias?

AI learns from existing data, including its one-sided perspectives — so biased inputs produce biased outputs, and a loss of multi-perspectivity.

A model has no view of the world independent of its training data; it absorbs whatever is in that data, including the prejudices and blind spots baked into it. If the data over-represents one perspective, the model reproduces and amplifies that skew — "garbage in, garbage out." The deeper cost Budelacci names is a loss of multi-perspectivity: the plurality of viewpoints that critical thinking depends on gets flattened into whatever the majority of the data happened to encode. So an AI's apparent neutrality is an illusion; it inherits, and can entrench, the biases of its source material.