The Friendly CHRO

The Human Side of AI-Powered HR

AI Hallucination: Why It Happens and How We Can Fix It

Written by

Sharad

AI tools like ChatGPT, Gemini, and Claude are transforming how we work, learn, and create. But sometimes, they get things very wrong—confidently stating false facts, making up events, or even inventing entirely fake sources.

AI Hallucination: Why It Happens and How We Can Fix It
AI Hallucination: Why It Happens and How We Can Fix It

This phenomenon is called AI hallucination, and it’s one of the biggest challenges in artificial intelligence today.

In this post, we break down:

  • What AI hallucination really is (with real-world examples)
  • Why it happens (in simple, non-technical terms)
  • How companies are trying to fix it
  • What you can do to avoid being misled

This article is part of a series demystifying Gen AI and its applications for the world of work. You can read the earlier articles – a basic guide to LLMs, prompt engineering vs context engineering, AI bots vs agents, intro to RLHF and how Gen AI models are trained and about how AI can predict human cognition.

What Exactly is AI Hallucination?

Imagine asking a friend for a restaurant recommendation, and instead of suggesting a real place, they invent one—complete with a fake menu and glowing (but imaginary) reviews.

That’s essentially what AI hallucination is:

  • AI makes up information that sounds plausible but isn’t real.
  • It doesn’t do this intentionally—it’s not “lying.” It’s just guessing based on patterns.
  • The more confident the AI sounds, the more dangerous this can be.

Real-Life Examples of AI Hallucination

1. The Lawyer Who Cited Fake Cases

In 2023, a New York lawyer used ChatGPT for legal research. The AI invented six completely fake court cases, complete with made-up quotes and references. The lawyer didn’t verify them and submitted them to court—leading to fines and embarrassment.

2. Google’s AI Bot Claimed It Invented a New Law

During a demo, Google’s AI (Bard) was asked about new discoveries from the James Webb Space Telescope. It falsely claimed that the telescope took the “first-ever pictures of an exoplanet outside our solar system”—when in fact, that happened years earlier.

3. AI-Generated Medical Advice Gone Wrong

A study found that when asked medical questions, some AI models suggested dangerous, unproven treatments—like recommending a fake drug for a real condition.

Why Do AI Models Hallucinate?

AI doesn’t “think” like humans do. It predicts words based on patterns in its training data—not on real-world knowledge.

Here’s why hallucinations happen:

1. AI is a “Supercharged Guessor,” Not a Fact-Checker

  • Large Language Models (LLMs) like ChatGPT don’t “know” facts—they predict the next word in a sentence.
  • If the best guess is wrong, the AI will still say it confidently.

Example:
If you ask, “Who invented the telephone in 2025?”
A human would say, “That doesn’t make sense—the telephone was invented in 1876.”
But an AI might make up a fake inventor because it’s trying to complete the sentence, not fact-check.

2. Gaps in Training Data

  • If the AI wasn’t trained on enough accurate information, it fills in the blanks with guesses.
  • This is especially common for niche topics, recent events, or very specific questions.

3. Over-Optimization for “Helpfulness”

  • AI is designed to give complete, smooth-sounding answers—even when it’s unsure.
  • So instead of saying “I don’t know,” it often makes something up to seem helpful.

4. No Built-In Fact-Checking (Yet)

  • Unlike a search engine (which pulls from real websites), AI generates text from scratch.
  • Without external verification, it can’t tell if what it’s saying is true.

How Are Companies Fixing AI Hallucination?

Tech companies know this is a major issue, and they’re working on solutions. Here are some approaches:

1. Improved Training with High-Quality Data

  • Companies are feeding AI more accurate, well-sourced data to reduce gaps.
  • Some models now use human-reviewed answers to learn correctness.

2. Reinforcement Learning from Human Feedback (RLHF)

  • Humans rank AI responses (good vs. bad), helping the AI learn which answers are reliable.
  • Over time, this reduces made-up responses.

3. Fact-Checking with External Sources

  • Some AI models (like Microsoft’s Copilot) cross-check answers against trusted sources (Wikipedia, news sites, etc.).
  • If the AI isn’t sure, it might say: “I couldn’t verify this—please double-check.”

4. Adding Uncertainty Warnings

  • Instead of sounding 100% confident, newer AI models are being trained to say:
  • “I’m not entirely sure, but…”
  • “This might need fact-checking…”

5. Hybrid AI + Search Models

  • Some tools (like Perplexity AI) combine generative AI with live web searches to provide sources.
  • This helps avoid pure “made-up” answers.

How Can You Avoid AI Hallucinations?

Until AI gets better at fact-checking itself, here’s how you can stay safe:

✔ Always Verify Critical Information

  • If an AI gives you a fact, name, or statistic—Google it before trusting it.
  • This is especially important for medical, legal, or financial advice.

✔ Ask for Sources

  • Some AI tools (like Copilot) provide links to where they got their info.
  • If the AI can’t provide a source, be skeptical.

✔ Use AI for Brainstorming, Not Final Answers

  • Great for: “Give me ideas for blog topics.”
  • Risky for: “Tell me the exact legal requirements for my business.”

✔ Report Wrong Answers

  • Many AI platforms let you flag incorrect responses, helping improve the system.

The Future of AI Hallucination

AI is improving fast, but hallucinations won’t disappear overnight. The next generation of models will likely:

  • Integrate real-time fact-checking more seamlessly.
  • Be more transparent about uncertainty.
  • Work alongside search engines to reduce made-up answers.

For now, the best approach is: Trust, but verify.

External reading:

When AI Gets It Wrong: Addressing AI Hallucinations and Bias

What are AI hallucinations

Check out these popular stories:

Discover more from The Friendly CHRO

Subscribe now to keep reading and get access to the full archive.

Continue reading