Stop Calling AI Errors “Hallucinations” — It’s Time for Clear Accountability

In the race to define artificial intelligence, language matters. One of the most common—and problematic—terms used today is - AI "hallucinates." This phrase, is often used to describe when AI generates factually incorrect or fabricated outputs, may sound catchy, but it's deeply misleading and potentially harmful.

Unlike humans, AI systems don’t possess consciousness, perception, or intent. When an AI tool like ChatGPT produces false information, it's not “hallucinating”—it’s simply generating text based on patterns in its training data without understanding truth or context. Framing this behaviour as hallucination anthropomorphises the system, making it seem more sentient or self-aware than it is.

This mischaracterization creates confusion about AI's capabilities and limitations. It may lead users to place undue trust in AI outputs or, conversely, fear AI systems as unpredictable or sentient entities. Both outcomes distort public understanding and hinder responsible adoption.

Instead, we should use precise terms like "output errors," "inaccuracies," or "model failures." These emphasize the mechanical, non-conscious nature of AI and keep the focus on improving system reliability rather than sensationalizing its flaws.

As AI becomes more integrated into society, getting the language right isn’t just semantics—it’s critical to safe and ethical deployment.