Which statement about biases in AI-generated text is most accurate?

Test your Media and Information Literacy skills. Prepare with multiple choice questions and detailed explanations to enhance your understanding. Ace your exam with our comprehensive study resources!

Bias in AI-generated text comes from several intertwined factors, not from a single source. The training data carry existing biases—stereotypes, under- or overrepresentation of groups, labeling quirks—that the model can reproduce in its outputs. How you frame a task or prompt also shapes the result, because the instructions and examples you provide guide what the model thinks it should produce. At the same time, the model tends to generalize patterns it learned during training to new contexts, which can spread biased or inaccurate assumptions beyond where they originally appeared. So the statement that best captures what’s happening points to training data biases, framing, and overgeneralization as the main sources of bias in AI-generated text.

It’s not about algorithmic complexity alone—the complexity of the model doesn’t erase bias and can even amplify it. Large datasets don’t guarantee bias-free outputs, because representation issues and skewed patterns can persist regardless of volume. And bias is a real concern for text generation because it affects fairness, accuracy, and trust in what the AI produces.