How can bias in machine learning models impact computer vision tasks, and what steps can be taken to reduce this bias?