4 min read AI

Hallucination is Human: Are AI's "Fails" Just a Mirror?

The moment AI “hallucinates” and produces incorrect, fabricated, or nonsensical information, it becomes headline news. We immediately label it as untrustworthy, fundamentally flawed, and a clear reason to keep it at arm’s length. “See!” we exclaim, “It’s just making things up!”

But let’s be honest with ourselves:

Isn’t “making things up” an inherently human trait?

The Human Condition: Flawed, Forgetting, Fabricating

Our brains are not perfect, unbiased databases. They are incredible storytelling machines, constantly filling in gaps, making assumptions, and yes, sometimes outright fabricating details to create a coherent narrative.

Consider these everyday “human hallucinations”:

  • Misremembering Events: How often do two people recall the same shared experience with wildly different details? Our memories are reconstructive, not recordings.

  • Quoting Incorrectly: We often misattribute quotes, conflate facts, or blend different pieces of information into a convenient but inaccurate summary.

  • Confabulation: In certain neurological conditions, individuals genuinely create and believe false memories without any intent to deceive. It’s the brain filling in the blanks, in the same way it does when our eyes move.

  • Overconfidence in Expertise: Experts in a field can sometimes make definitive statements that turn out to be incorrect, based on outdated information or personal bias, simply because their confidence overrides their accuracy.

Are these not forms of “hallucination”? Yet, we don’t dismiss the entire human intellect because of them. We understand they are part of the messy, brilliant process of human cognition.

AI’s Hallucinations: A Different Flavour, Not a Different Sin

When an AI hallucinates, it’s often because it’s doing precisely what it was trained to do: predict the next most plausible token based on patterns in vast datasets. If those datasets contain biases, gaps, or conflicting information, or if the prompt is ambiguous, the AI will plausibly fill the void. It’s not malicious; it’s statistical.

The key difference isn’t that AI hallucinates and humans don’t. It’s that AI’s hallucinations can be more systematic, or appear in contexts where we expect objective truth. This is closely tied to the median knowledge problem — LLMs give you the aggregate of their training data, and sometimes that aggregate is confidently wrong.

The Car Accident Analogy

Think about autonomous vehicles. When a self-driving car has an accident, it’s a major news story. The immediate conclusion is often: “AI cars are too dangerous!”

Yet, human drivers have accidents far, far more often. The WHO reports that 1.19 million people die annually, of which 94% are crashes involving human error. We accept this level of human fallibility as part of life, but demand absolute, unerring perfection from a machine.

Just because an AI gets something wrong, or makes up a detail, doesn’t mean everything it produces is invalid.

Our Role: Responsible Collaboration, Not Blind Trust or Dismissal

Understanding that “hallucination is human” doesn’t mean we should accept AI’s inaccuracies without question. Instead, it shifts our perspective:

  1. Context is King: Evaluate AI outputs based on the stakes. A creative brainstorming session is different from a legal brief.

  2. Verify, Verify, Verify: Just as you wouldn’t blindly trust every statistic or claim from a human source, you must verify critical information from AI.

  3. Design for Robustness: Developers must continue to build safeguards, retrieval-augmented generation (RAG) systems, and better grounding mechanisms to minimise factual errors.

  4. Embrace the “Messiness”: We should see AI as a powerful, albeit imperfect, collaborator. Its occasional “hallucinations” can even spark new ideas, much like a human brainstorming partner might say something outlandish that triggers a brilliant thought.

The pursuit of perfect, un-hallucinating AI is a noble one, but perhaps unrealistic, given the nature of intelligence itself. By acknowledging that this tendency is deeply embedded in both human and artificial cognition, we can foster a more realistic, productive, and ultimately more effective partnership with AI.

What are your thoughts? Should we be more forgiving of AI’s “hallucinations” given our own human tendencies?

Hashtags: #ArtificialIntelligence #AIEthics #FutureOfWork #Humanity #Cognition #Innovation #Leadership #ResponsibleAI