That Mike HamiltonHow to tell when AI is wrong
AI's biggest flaw isn't that it makes mistakes. It's that it makes them in the same calm, confident voice it uses for the truth. Here's how to catch it.
The most important thing to understand about AI is this: it doesn't look up facts. It predicts words. When you ask a question, it's assembling the most likely-sounding answer based on patterns it learned — not checking a reference book. Most of the time that lands on the truth, because the truth is what usually gets written down. But when it doesn't know, it doesn't stop. It fills the gap with something that sounds right.
There's a name for this: a hallucination. It's when AI states something false as if it were solid fact — a made-up statistic, a quote nobody said, a court case that never happened, a wrong date delivered with total confidence.
Why it happens (the 20-second version)
Think of AI as the world's most well-read improv actor. Ask it anything and it will give you a fluent, in-character answer — because its job is to keep the scene going, not to say "I don't actually know." It's not lying; it has no idea it's wrong. That's exactly what makes it tricky: there's no nervous tell, no hesitation. The false answer looks identical to the true one.
The one habit that protects you
You don't need to fact-check everything. You need to fact-check the right things. Watch for what I call the risky specifics:
- Numbers — statistics, prices, dates, measurements.
- Names — people, companies, products, places.
- Quotes — anything in quotation marks attributed to someone.
- Citations — studies, laws, court cases, book titles, links.
When an answer hinges on one of those, verify it before you lean on it. Two easy ways: ask the AI "What's your source for that?" and actually check the source, or drop the claim into a normal web search. Thirty seconds of checking beats forwarding a made-up statistic to your whole team.
Trust AI with your thinking. Verify it with your facts.
When to be extra skeptical
Turn the dial up when the topic is recent (AI's knowledge has a cutoff and it may not know it), when it's oddly precise ("in a 2019 study of 1,247 participants…" is a classic hallucination shape), or when the stakes are high — anything medical, legal, or financial. For those, confirm with a trusted source or a qualified human. Full stop.
The bottom line
AI is a brilliant first-draft machine and a lousy final authority. Use it to move faster, then keep your hand on the wheel for the facts that matter. Do that, and the confident-wrong problem stops being a trap and starts being a thing you just… handle.