Loading field map
Every mahjong discard is a gamble. The player sees only thirteen of the 136 tiles, knows nothing about the opponents' concealed hands, and must decide in seconds whether to chase a high-scoring combination, race toward a cheap win, or abandon offense entirely to avoid dealing in. The history of mahjong game theory is the history of how players and analysts have tried to make that gamble less blind. Over the past century, five distinct methodological schools have emerged, each offering a different answer to the same question: given what you know and what you cannot know, which tile should you throw away?
The earliest systematic approach to mahjong strategy was not written down in textbooks. It lived in the heads of experienced players who had logged thousands of hands. The Classical Intuitive School treated discard choice as a pattern-recognition skill: a seasoned player could "feel" which tiles were dangerous because they had seen similar situations before. Heuristics such as "avoid discarding the same suit as an opponent's recent discard" or "hold pairs of honor tiles because they are hard to draw" were passed from teacher to student. This school had no formal probability calculations and no explicit scoring models. Its strength was speed—a practiced player could make reasonable decisions in seconds—but its weakness was inconsistency. Two intuitive players facing the same hand might reach opposite conclusions, and neither could explain why. The Classical Intuitive School never disappeared entirely; it survives today as the background judgment that even mathematically trained players rely on when time is short or data is thin. But by the mid-twentieth century, a growing number of players wanted something more reliable than intuition.
The Probability-Based School introduced a new question: instead of asking "what feels right?", it asked "what is the expected value of this discard?" This school treated mahjong as a problem of decision under uncertainty, borrowing tools from statistics and early game theory. Its core method was tile-out probability: given the tiles already visible (discards, melds, and one's own hand), what is the chance that a particular tile remains in the wall or is held by an opponent? Players learned to compute the number of "live" copies of each tile and to estimate the likelihood that a given discard would complete an opponent's hand. The school also introduced expected-value reasoning for hand valuation: a hand that wins quickly for a small score might be preferable to a slow hand with a larger potential score, once the risk of dealing in is factored in. The Probability-Based School did not reject intuition entirely—it could not, because many probabilities were too complex to calculate at the table—but it narrowed the role of intuition to situations where calculation was impossible. It also laid the groundwork for the two schools that followed, each of which absorbed its probabilistic methods while adding new priorities.
By the 1990s, the Probability-Based School had shown that mahjong decisions could be analyzed mathematically, but it had not settled which mathematical priorities should dominate. Two new schools emerged in parallel, each taking the probabilistic toolkit in a different direction.
The Defense and Push-Fold Strategy School focused on the binary choice at the heart of every turn: should you push toward your own win or fold into a purely defensive posture? This school borrowed the push-fold concept from poker tournament strategy and applied it to mahjong. Its key insight was that the cost of dealing in (paying a large point penalty to the winner) often outweighs the benefit of completing a marginal hand. The Defense School developed explicit thresholds: if your hand is below a certain expected value or if an opponent shows strong signs of readiness (tenpai), the correct play is to discard only safe tiles—tiles that cannot possibly complete an opponent's hand. This school's methods include reading opponent discards for suji (safe intervals) and kabe (blocked suits), and it treats every discard as a risk-management decision. Its recommendations often diverge sharply from those of the Tile-Efficiency School: where a defense-oriented player sees a dangerous tile and folds, a speed-oriented player might see an opportunity to advance their hand and push ahead.
The Tile-Efficiency and Speed Play School, emerging at roughly the same time, argued that the primary goal of the early and middle game is to reach tenpai as quickly as possible, regardless of score. Its central method is shanten counting (measuring how many tile draws are needed to reach tenpai) and ukeire counting (counting the number of distinct tiles that improve the hand). The school introduced concepts such as five-block theory (building a hand around five tile groups rather than chasing random pairs) and iishanten peak theory (recognizing that the hand is most flexible one step away from tenpai). Where the Defense School sees risk, the Tile-Efficiency School sees opportunity: a hand that reaches tenpai early can win before opponents have time to build dangerous hands, and even a cheap win is better than no win. The two schools coexist in productive tension. In concrete terms, they disagree most sharply on hands that are one or two steps from tenpai but require discarding a tile that is moderately dangerous. The Defense School says fold; the Tile-Efficiency School says push. Professional players today must decide which school's logic applies to the specific table situation, and many have developed hybrid approaches that switch between the two depending on score position, round number, and opponent tendencies.
The most recent school did not emerge from human reasoning at all. The AI-Assisted Mahjong Analysis School uses machine learning, Monte Carlo simulation, and game-tree search to evaluate positions that are too complex for human calculation. Early mahjong AI engines, such as the Japanese "Naga" and "Suphx" systems, were trained on millions of expert games and learned to predict the expected value of every possible discard with far greater accuracy than any human. The AI School has done more than just produce stronger play; it has acted as an empirical referee in the long-standing dispute between the Defense and Tile-Efficiency schools. By running millions of simulations, AI analysis has shown that the Tile-Efficiency School's advice is generally correct in the early game—speed matters more than safety—but that the Defense School's caution becomes dominant once an opponent reaches tenpai. The AI has also revealed that some intuitive heuristics from the Classical Intuitive School are surprisingly robust (such as the value of holding pairs of honor tiles), while others are systematically wrong (such as overvaluing closed hands at the expense of speed). The AI School has not replaced human judgment; rather, it has provided a tool that allows players to test their assumptions and calibrate their intuition. Today, top players routinely study AI output to refine their decision-making, but they still rely on the older frameworks to make real-time choices when they cannot run a simulation in their heads.
The five schools are not a linear succession. The Classical Intuitive School persists as a fast heuristic layer; the Probability-Based School provides the underlying mathematics that the later schools inherit; the Defense and Tile-Efficiency schools remain in active competition; and the AI School serves as an external validator. What the leading frameworks agree on is that mahjong is a game of incomplete information where expected value is the correct normative standard, and that no single heuristic works in every situation. What they disagree on is how to estimate expected value under time pressure and which risks are worth taking. The most active debate today concerns the weight that should be given to AI recommendations: some players argue that engine output should override human judgment entirely, while others maintain that AI analysis is only as good as the assumptions built into its simulation parameters. Another open question is whether the Tile-Efficiency School's emphasis on speed is optimal in all rule variants, or whether it is specific to Japanese riichi mahjong with its small hand values and fast-paced structure. These debates ensure that mahjong game theory remains a living field, not a settled doctrine. The student who understands all five schools will be better equipped to decide, at every turn, which tile to throw away.