Stop Saying AI "Hallucinates"—It’s Just Wrong

Before I get on my soapbox and rant a little, I want it known, I have nothing against AI, I use it almost everyday since it's integrated into tools I use for work and personal. Yet I think we talk about AI differently that we would any other piece of technology, for the worse. I started this rant in my head after hearing a security talk where the presenter was telling the story of having an AI model tested for security. A participant in the test, thought that they got the AI to share information from documents that they shouldn't have been able to. Now for the talk this was seen as a win, the presenter called the AI so advanced that instead of giving the tester information, it "hallucinated" information instead of disclosing information. No, it didn't hallucinate, it was wrong and gave wrong information. If I go to my calculator and type 2+2= and it spits out 32,456, I wouldn't say it hallucinated an answer....I'd say it was broken (shows me for buying my calculators from a discount store)

Let me make this simplifier, imagine you’re at a restaurant, and the waiter brings you a steak when you ordered a salad. When you complain, the manager explains, “Our kitchen had a little culinary hallucination.” You’d probably stare at them like they were auditioning for a role in The Twilight Zone. The steak isn’t some imaginative leap—it’s a mistake.

The same principle applies to AI. Yet, in the tech world, when artificial intelligence spits out something wildly incorrect, we don’t call it what it is. We soften it, dress it up, and call it a "hallucination." Sounds quirky, even charming, doesn’t it? Like the AI is a misunderstood artist painting abstract visions of data. But let’s cut the spin: when AI "hallucinates," it’s wrong. Sometimes spectacularly so.

Euphemisms vs. Accountability

Why do we use words like "hallucinate"? Well, it’s more palatable than saying, “Oops, our cutting-edge, world-changing technology just barfed up nonsense.” By framing errors as hallucinations, tech companies and AI enthusiasts distance themselves from the messy reality: AI is fallible.

Sure, "hallucinate" sounds cool and sci-fi, but it also implies something creative or accidental rather than flawed. It lets us avoid hard conversations about accountability, limitations, and—let’s be honest—how often the tech isn’t ready for prime time.

AI hallucinations aren’t charming; they’re problematic. Picture this: You’re using an AI-powered assistant to draft an email or summarize a document, and it confidently fabricates facts. It doesn’t know it’s wrong because, spoiler alert, it doesn’t know anything. It’s not sentient or whimsical—it’s a super-speedy autocomplete machine cobbling together patterns from the data it’s been fed.

Let’s ditch the romanticism. The AI isn’t “hallucinating”—it’s malfunctioning. Just like your GPS isn’t “daydreaming” when it tells you to turn left into a river.

The Danger of Sugarcoating

Euphemisms like "hallucinate" don’t just muddy the waters—they actively harm our understanding of AI. When we talk about AI errors in flowery terms, we risk downplaying their impact. In areas like healthcare, law, or finance, these “hallucinations” can lead to serious consequences. Misinformation, misdiagnoses, or legal mishaps aren’t just little "oopsies" they’re high-stakes failures.

If we don’t call out these issues for what they are—errors, flaws, bugs—we’re letting AI off the hook. Worse, we’re letting the companies behind it sidestep responsibility.

Why Plain Language Matters

Calling an AI failure what it is—an error—doesn’t just make the tech industry sound more honest (and less like a Black Mirror episode). It also helps set realistic expectations for users.

AI isn’t magic, and pretending it is does no one any favors. Users need to understand the limitations of the tools they’re using, whether it’s ChatGPT drafting their emails or a recommendation system suggesting what to watch next. If the tech fails, they need to know it’s not because the AI “got creative” but because it has inherent limitations.

Let’s Get Real (and Stop the Spin)

Here’s a thought experiment: Instead of romanticizing AI errors, let’s approach them like we would any other product flaw. When your toaster burns your bread, no one says it’s “having a heated epiphany.” It’s just broken. Treat AI the same way.

And while we’re at it, let’s hold the creators of these systems accountable. If an AI bot is spouting falsehoods, that’s not just the bot’s “quirk”—it’s a problem with how it was designed, trained, or implemented. Accountability needs to rest squarely on the shoulders of the developers, not the machine.