When the opponent is a thinker
Most probability problems involve chance that does not respond to you: dice, shuffled cards, loot tables. Competitive games are different. Your opponent adapts, notices patterns and tries to exploit them. Game theory studies decisions in this kind of setting, and probability enters because the best way to act is sometimes to be deliberately unpredictable.
This article introduces the idea at a simple level. It is about understanding strategic choice in games, not about beating any gambling game; where the opponent is a casino with a built-in edge, no mixing of choices removes it.
Pure and mixed strategies
A pure strategy means always making the same choice in a given situation. A mixed strategy means choosing among options according to set probabilities, such as playing one option 60% of the time and another 40%.
Mixed strategies matter when a pure strategy can be countered. If your opponent can learn what you will do, they can pick the best response to it. Randomising removes their ability to do that.
The rock-paper-scissors example
Rock-paper-scissors is the standard illustration. Rock beats scissors, scissors beats paper, paper beats rock.
Suppose you favour rock and play it half the time, paper 25% and scissors 25%. If your opponent notices and responds by always playing paper:
- Paper beats your rock: they win 50% of rounds.
- Paper ties your paper: 25% ties.
- Paper loses to your scissors: they lose 25% of rounds.
Their average result per round is 0.50 − 0.25 = +0.25, which is a quarter of a point in their favour. Your habit has a cost, because it is exploitable.
Now suppose you play each option with probability 1/3. Whatever the opponent plays, they win a third of rounds, lose a third and tie a third. Their average is zero, and there is nothing they can do to improve it. That is the point of a mixed strategy: it makes you unexploitable, even if it does not make you a winner.
Why equal mixing is not always right
Rock-paper-scissors is symmetric, so an even split works. Most real matchups are not symmetric. In general, the right mix is one that leaves the opponent indifferent between their options, so that no single response is better against it. The exact proportions depend on the payoffs in the specific game, which means you cannot read them off without knowing those numbers.
Take a simple illustration with invented numbers. An attacker chooses between two routes, and a defender guards one of them. If the defender guards the route the attacker uses, the attack fails; otherwise it succeeds. If the routes are equally valuable, an even 50/50 mix for both sides is hard to exploit. If one route is more rewarding for the attacker, the best mix shifts, and the defender should guard the more rewarding route more often. Exact answers require the exact payoffs.
Randomising in real competitive games
Many competitive situations contain this structure:
- Choosing between attacking and defending when the opponent is guessing.
- Selecting among similar openings, so opponents cannot prepare specifically.
- Mixing up timing of a surprise action.
In practice, no one randomises with a coin. Humans are poor at it, tending to switch too often or avoid repeating a choice, which is itself a pattern. A related misunderstanding is the law of averages: after a run of one choice, people feel that another is due, even though independent random choices have no memory. Opponents sometimes exploit that.
One method that is sometimes suggested is to use something irrelevant and hard to guess, such as the final digit of a clock, to decide. This is a reasonable idea for casual play, though how well it works depends on the game and the opponent.
Limits of the theory
- Opponents are not perfect. A mixed strategy guarantees you cannot be exploited by a perfect opponent. Against a weaker opponent with a clear habit, deviating to exploit them may earn more, at the price of becoming exploitable yourself.
- Real payoffs are rarely known precisely. Estimates are uncertain, so the exact proportions are approximate.
- Randomness does not rescue weak play. An unpredictable mix of poor options is still poor.
- Skill still matters. Randomising helps decide among choices that are roughly equal in value.
This is also why a pure luck versus skill description misses what happens in many matchups: a good mix is itself a skilled choice.
Connection to probability
The mathematics behind this is ordinary probability. The average result of a mixed strategy against a given opponent response is an expected value: multiply each outcome by its chance and add. A mix is good when it leaves no response with a positive expected value against you. Over many rounds, results settle near those averages thanks to the law of large numbers, but any short sequence can look different.
Practical takeaways
- If an opponent can predict you, they can exploit you.
- Mixing choices protects against that, at the cost of giving up the chance to exploit a weaker opponent.
- The right proportions depend on the payoffs, not on a universal rule.
- People randomise badly, so deliberate variation helps, but no approach guarantees wins.
Competitive success also depends on mechanics, knowledge and consistency. Game theory describes a piece of strategic reasoning, not a formula for winning.