Rock Paper Scissors vs AI

Rock paper scissors against an opponent that learns

This is not the random opponent you played as a child. It watches every throw you make, builds a model of your recent patterns, and predicts what you will do next. Against pure randomness you would win a third of the time by definition. Against this, most people win considerably less, and the reason is that humans are extremely bad at being random.

How to play

1. Pick rock, paper or scissors.
2. The opponent throws at the same time, using its prediction of you.
3. Your win rate is tracked as you go. Play a long run rather than a few rounds.

Why you lose to a pattern reader

Human randomness has a very specific signature. We under-repeat, meaning we avoid throwing the same thing twice in a row far more than chance would, because a repeat feels less random than an alternation. We also cycle in short loops without noticing, and we react predictably to the previous result.

That last one is the strongest tell, and it has a name in competitive circles: win-stay, lose-shift. After a win people tend to repeat what worked. After a loss they tend to switch, and specifically they tend to switch to the throw that would have beaten what just beat them. A study of a large rock paper scissors tournament in China found exactly this conditional-response pattern rather than the mixed strategy game theory says is optimal.

How to actually beat it

• Play the opposite of your instinct after a loss. If you just lost with rock, your instinct says paper. That is precisely what a pattern reader expects.
• Deliberately repeat. Throwing the same thing three times in a row feels wrong and is statistically normal. Doing it breaks the model badly.
• Use an external source. Second hand on a clock, the character count of the last word you read, anything that is not you. This is the only genuinely reliable method.
• Do not think in threes. People planning ahead almost always plan a rotation, and a rotation is the easiest pattern in the world to detect.
• Play long. Over ten rounds anything can happen. Over a hundred the model either has you or it does not, and that is the real result.

The game theory footnote

The optimal strategy in rock paper scissors is to throw each option exactly one third of the time at random. That guarantees you cannot be exploited, and it also guarantees you cannot win more than a third, so it draws in the long run. Any attempt to beat an opponent means deviating from it, which makes you readable in turn. That tension is why the game survives at all as a contest rather than a coin toss.

Related

Reaction Time and Reflex Tap test speed rather than prediction, and Decide For Me will settle a two-way choice without any of this. All free, in the browser, nothing stored.

Watch the AI think, and see how random you really are

Every round shows what the model expected you to throw, how confident it was, and how often those guesses have landed. Alongside it sits a randomness score built from two things: how evenly you spread your throws, and how flat your transitions are from one move to the next. Score above eighty and you are genuinely hard to model. Drop under forty and you are running on habit, which is exactly what the chain feeds on. A running tape of your last twelve rounds and a best win streak sit next to the board, so you can see a losing pattern forming before the scoreline does.

Each round runs the full ritual: rock, paper, scissors, then the throw, with a tick under every beat and a whoosh on the reveal. Wins land with a rising chime and a burst of confetti, a third consecutive win upgrades that to a full fanfare, and losses land with a low thud and a jolt of the board. All of it is generated in the browser, so nothing is downloaded, and a single toggle in the tray turns it off with your choice remembered for next time.