Discriminatory patterns that emerge when AI models process multiple input types (text, audio, images) together.