What are the risks associated with biased or incomplete training data in AI?

• A. They can lead to hallucinations and other inaccuracies in AI outputs.
• B. They increase model throughput.
• C. They reduce data volume.
• D. They always improve fairness.

Answer: (A)

Biased or incomplete training data distort what the model learns, so it generalizes in ways that reflect those biases and gaps. When the model encounters new inputs, it tries to produce outputs based on patterns it saw in training, which can lead to plausible-sounding but incorrect results—hallucinations. Missing or skewed coverage also means the model may misinterpret or misrepresent scenarios it hasn’t seen enough of, resulting in inaccuracies.

This isn’t about processing speed or data volume. Throughput measures how fast the system runs, not the quality of what it was trained on. Data volume isn’t the core issue here either—the problem is the biased or incomplete nature of the data. And biased data does not improve fairness; in fact, it often makes outcomes less fair.