AI Deception Tutorial

Last Friday, my friend Sarah was convinced she'd spotted all the AI answers in our game. She had a whole theory about sentence length and punctuation patterns. Final score? 2 out of 7 correct. She'd been confidently voting for human-written answers as "obviously AI" the entire time.

That's AI Deception mode in a nutshell—it messes with your assumptions about how AI "should" sound.

How This Mode Actually Works

Everyone gets the same trivia question—something like "What year did the Berlin Wall fall?" or "What's the chemical symbol for gold?"

Players type their answers. Meanwhile, AI also generates 2-3 answers that could be plausible (sometimes correct, sometimes confidently wrong). All answers get shuffled together. Your job? Vote on which ones came from AI.

What I Learned After Playing 50+ Games

Forget everything you think you know about "how AI writes." Modern AI can sound casual, make typos, use slang—basically everything people say is a telltale sign. Here's what actually works:

The "Confidence Without Personality" Tell

AI answers tend to be assertive but bland. Check this out:

The Wikipedia Echo

Sometimes AI pulls phrasing directly from its training data. If an answer sounds like it was copied from an encyclopedia entry, trust your gut:

When Being Wrong Feels Too Right

Humans get stuff wrong in weird ways. We confuse decades, mix up similar-sounding names, or half-remember stuff from high school. AI, when it's wrong, tends to be wrong in... let's call it "algorithmically plausible" ways.

Example question: "Who wrote Romeo and Juliet?"

• Human wrong answer: "uhh Marlowe? or was that Macbeth"
• AI wrong answer: "Christopher Marlowe wrote Romeo and Juliet in 1594."

See the difference? Humans question themselves. AI commits to the bit.

The Length Trap (And Why It Fails)

Early strategy guides said "check answer length—AI writes longer answers." Not anymore. I've seen AI write single-word answers and humans write paragraphs. Don't rely on this.

Actually Useful Detection Patterns

Hard Mode Strategy: When AI Gets Sneaky

In higher difficulty settings, AI learns from previous rounds and mimics the group's writing style. This is where things get psychological.

Counter-strategy: Look for consistency. If someone (or something) writes in basically the same style every round—same length, same vibe, same punctuation habits—that's probably AI. Humans vary naturally.

I watched a player named "Dave" write three rounds in a row with eerily similar phrasing: "I believe it's X," "I think it's Y," "It's likely Z." Turns out Dave was AI. Real Dave had left after round two to make popcorn.

Common Mistakes I See Every Game

Mistake 1: Voting based on whether the answer is correct. AI can be wrong. Humans can be right. You're judging the writing style, not the accuracy.

Mistake 2: Thinking typos automatically mean human. AI in hard mode includes deliberate typos. Look at where the typos occur—humans typo differently than AI trying to simulate typos.

Mistake 3: Overthinking it. If your gut says "this sounds like a robot," it probably is. Our brains are pretty good at this if we stop second-guessing ourselves.

Quick Practice Round

Question: "What's the capital of Australia?"

Final Thoughts From Someone Who's Played Too Many Rounds

The meta is constantly shifting as AI gets better. What worked three months ago (checking for perfect grammar) barely works now. What works today (looking for unnatural hedging) might not work next month.

Best advice? Play a bunch of rounds, make terrible calls, learn from what your friends caught that you missed, and develop your own intuition. This mode is less about memorizing rules and more about developing a sixth sense for robot vibes.

And when in doubt, remember: AI is really good at sounding informative. Humans are really good at sounding like we're winging it because we usually are.