Human vs AI Rock Paper Scissors Tournament — Who Won?

Event: "The 10-Million-Won One-Shot Rock Paper Scissors Tournament" Host: AI startup Paya Venue: Sogang University Date: August 2017 Entrants: • Human division: around 100 members of the public • AI division: 19 algorithm teams • Format: divisional preliminaries → grand final Advising professor Koo Myung-wan (Sogang University, Department of Computer Engineering) called it "the first Rock Paper Scissors tournament in Korea to make use of artificial intelligence techniques."

In the grand final, the overall champion was Jo Min-kyung (46), from the human division. She defeated the winner of the AI division and took home the 10 million won. Jo's secret to victory: "I studied the preliminary records of the AI contestants and tried hard to figure out which throws they favored. Sticking to my own pattern is what carried me all the way to a come-from-behind win." In other words, Jo "analyzed" the AI, while the AI simply didn't have enough time to learn her pattern — and lost.

AI dominates at chess and Go, but Rock Paper Scissors is different. 1. No information — RPS is a simultaneous game where you can't see your opponent's hand in advance 2. Limited training data — a single round or short match makes pattern learning difficult 3. True randomness — if a human uses an external random source (like a clock's second hand), AI can only win about 33.3% of the time statistically Conversely, where AI is strong: • Fast pattern learning (over 1,000 rounds against the same opponent, it dominates) • Exploiting statistical biases (such as the fact that people rarely repeat the same throw)

Multiple academic studies reach the same conclusion. • Wang et al. (2014, China): a 360-person experiment — found a Win-Stay/Lose-Shift bias • Walker & Walker (1997): demonstrated a statistical bias in first throws • MIT research (2014): once play becomes "truly random," AI always sits at a 33.3% win rate Bottom line: AI wins by exploiting human patterns, but if a human deliberately breaks their pattern, AI can never beat them statistically.

The Gababo AI bot learns from your last N throws. In typical casual play, your pattern surfaces unconsciously, so the Gababo bot's win rate climbs to around 55-60% — but if you use an external random source, it drops below 50%. The surest way to beat the bot: apply an external rule based on a clock's second hand, such as 0-3-6-9 = rock, 1-4-7 = scissors, 2-5-8 = paper.

• Rock Paper Scissors mind games (an in-depth AI vs human analysis): /psychology • How to win at Rock Paper Scissors: /strategy • Play against the bot: /game/bot/single