# Go, the Board Game, From Zero > Learn Go (weiqi, baduk) from zero: the few rules that matter, liberties and capture, ko, life and death, scoring with Japanese rules, opening ideas, tactics like ladders and snapbacks, reading ahead, and the story of AlphaGo. --- # Go, the Board Game, From Zero Go has about a page of rules and more depth than any other classic board game. You place a stone, the opponent places a stone, and from that tiny loop come fights, sacrifices, and moves that strong players still argue about after centuries. Nobody hands beginners the way in: the rules look too simple to be a game, and then you lose your first ten games without seeing why. This guide gives you the way in. Each phase teaches one layer, shows it on small diagrams that were checked move by move, and ends with a quiz so you know it stuck. You learn the Japanese rules, because they are the classic ones and the easiest to explain on a board, and one section shows how Chinese-style counting compares so you can follow any game you meet. ## How to read this - **Brand new to the game?** Read in order and play small. Use a 9x9 board (a printed one, a real set, or any free online board) for every example. A 9x9 game takes about 10 minutes and teaches the same ideas as the full 19x19 board. - **Know the rules already?** Jump to [Life and Death](03-life-and-death.md) or [Opening, Shape, and Tesuji](05-opening-shape-and-tesuji.md). - **Here for the AI story?** Read [Reading Ahead, and How AlphaGo Changed Go](06-reading-ahead-and-alphago.md), then continue to [How Computers Play Games](/guides/how-computers-play-games). ## The phases 1. **[Stones, Liberties, and Capture](01-stones-liberties-and-capture.md)** - the board, how a stone lives or dies, atari, and the one move you may not play. 2. **[Ko, and Why It Exists](02-ko-and-why-it-exists.md)** - the shape that could repeat forever, and the rule and the fights that come from it. 3. **[Life and Death](03-life-and-death.md)** - why two eyes make a group uncapturable, false eyes, and seki. 4. **[Counting: Territory, Prisoners, and Komi](04-counting-territory-prisoners-and-komi.md)** - how a game ends and who wins, plus how area scoring compares. 5. **[Opening, Shape, and Tesuji](05-opening-shape-and-tesuji.md)** - why corners come first, efficient shape, and tactics you can verify by reading. 6. **[Reading Ahead, and How AlphaGo Changed Go](06-reading-ahead-and-alphago.md)** - how to think several moves deep, and how a program learned to beat a world champion. > Sibling guides in this category teach [Sudoku](/guides/sudoku-from-zero), [Checkers](/guides/checkers-from-zero), and [Chess](/guides/chess-from-zero). Reading ahead is the skill they all share. --- # Stones, Liberties, and Capture If you can tell whether a stone is about to be captured, you can play Go. That is the whole skill of this phase: count one number, the liberties, and know what happens when it reaches zero. Everything else in the game, from eyes to territory, is built on top of that count. ## The board and the loop Go is played on the **intersections** of a grid, not inside the squares. The full board is 19x19, but 13x13 and 9x9 boards exist for learning. Start on 9x9: games are short and the ideas are identical. ```text 1 2 3 4 5 6 7 8 9 1 . . . . . . . . . 2 . . . . . . . . . 3 . . + . . . + . . 4 . . . . . . . . . 5 . . . . + . . . . 6 . . . . . . . . . 7 . . + . . . + . . 8 . . . . . . . . . 9 . . . . . . . . . ``` The `+` marks are only landmarks printed on the board (called star points). Nothing about them is special in the rules. The loop of the game: 1. **Black plays first.** Players alternate, placing one stone of their color on an empty intersection. 2. **A stone never moves** after it is placed. It stays until it is captured. 3. **You may pass** instead of placing a stone. Two passes in a row end the game (Phase 4 covers the ending and counting). The goal is to surround more of the board than your opponent. Placing a stone does not capture anything by itself; capturing comes from the next idea. ## Liberties: a stone's breathing room A **liberty** is an empty point directly next to a stone, up, down, left, or right. Diagonals do not count. A stone with liberties is safe for now. A stone with none is captured. ```text 1 2 3 4 5 6 7 8 9 1 X . . . X . . . . 2 . . . . . . . . . 3 . . . . . . . . . 4 . . . . . . . . . 5 . . . . X . . . . 6 . . . . . . . . . 7 . . . . . . . . . 8 . . . . . . . . . 9 . . . . . . . . . ``` Count them. The stone in the middle of the board, at the center, has **4** liberties. The stone on the top edge (row 1, column 5) has **3**, because the edge is a wall, not an empty point. The stone in the corner (row 1, column 1) has **2**. That is the first lesson of position: edges and corners take liberties away, so stones there are easier to capture and need less to defend. ## Groups share liberties Stones of the same color that touch along a line (not diagonally) form a **group**. A group lives and dies as one unit and shares all its liberties. ```text . . . . . . . . . . X X . . . X . . . . . . . . . X . . . . . . . . . . ``` The two stones on the left form one group with **6** liberties together. The two stones on the right touch only at a corner, so they are not connected: each is its own group with 4 liberties, and the two groups share two of them. That diagonal gap is called a **cutting point**. A group that is connected cannot be split. Two stones that are only diagonal can be separated if the opponent plays in the gap, which is why strong players think about connections constantly. ## Capture and atari When the last liberty of a group is filled by the opponent, the whole group is removed from the board. Removed stones are called **prisoners** (they matter in Phase 4). A group with exactly **one** liberty is in **atari**. Atari is a Japanese word, from the verb for "to hit": the group will be captured next move unless its owner does something. Here White's stone has one liberty, marked `a`: ```text . . . . . . . X . . . X O a . . . X . . . . . . . ``` Black plays at `a` and the White stone is captured: ```text . . . . . . . X . . . X . X . . . X . . . . . . . ``` *What just happened:* Black filled White's last liberty, so the White stone was lifted off the board and kept as a prisoner. The point it occupied is empty again. Groups are captured whole. Here the two White stones share a single liberty at `a` along the top edge: ```text +----------- | . X X . . | X O O a . | . X X . . | . . . . . ``` After Black plays `a`, both stones go at once: ```text +----------- | . X X . . | X . . X . | . X X . . | . . . . . ``` ## The one illegal move: suicide You may not place a stone that leaves its own group with no liberties, unless that move captures something. Black cannot play at `a` here, because the stone would have no liberties and captures nothing (every White stone around it still has outside liberties): ```text . . . . . . . O . . . O a O . . . O . . . . . . . ``` But the same kind of move is legal when it captures. In this corner Black plays `a`, which has no liberty of its own, yet it takes the last liberty of both White stones and removes them: ```text +------- | a O X . | O X . . | X . . . ``` After the capture the Black stone has two liberties: ```text +------- | X . X . | . X . . | X . . . ``` The order matters: captures are resolved first, and only then is the new stone checked for liberties. That is why this move is legal. This rule ("suicide is not allowed") is the same in Japanese and most other rule sets, but be aware a few rule sets do allow it. We use the Japanese rule. ## Your turn: read the position Check yourself before moving on: ```quiz [ { "q": "A single stone sits in the exact corner of the board with nothing around it. How many liberties does it have?", "choices": ["1", "2", "3", "4"], "answer": 1, "explain": "A corner stone has only two neighbors on the board, so it has two liberties. The edges take the others away." }, { "q": "What does atari mean?", "choices": ["A group has exactly one liberty and will be captured next move if nothing changes", "A group was captured", "A move that is not allowed", "The end of the game"], "answer": 0, "explain": "Atari is a warning: one liberty left. It is not a capture yet." }, { "q": "You place a stone where it has no liberties, but doing so removes an opposing group that had its last liberty there. Is the move legal?", "choices": ["No, suicide is never allowed", "Yes, because the capture happens first and gives the new stone liberties", "Only on the first move of the game"], "answer": 1, "explain": "Captures are resolved before the new stone is checked. The empty points left by the captured stones become the new stone's liberties." } ] ``` ## Recap 1. Stones go on intersections, Black first, alternating; stones never move once placed, and you may pass. 2. A liberty is an empty point directly adjacent (not diagonal); edges and corners reduce liberties. 3. Connected stones form a group that shares liberties and is captured as a unit. 4. One liberty left means atari; filling the last liberty removes the group as prisoners. 5. A move that leaves your own group with no liberties is illegal, unless it captures something first. Next up, [Ko, and Why It Exists](02-ko-and-why-it-exists.md): what happens when two players can capture each other's stone forever. --- # Ko, and Why It Exists Sooner or later you capture a stone, look at the board, and find your opponent can capture your capturing stone right back. Then you could capture again. And again. That loop has a name, ko, and one rule keeps it from running forever. The rule is short, but the fights it creates are some of the sharpest in Go. ## The loop that would never end Ko is the Japanese word for this situation, and it is the standard Go term everywhere now. It happens in a very specific shape. White's single stone is in atari, and the point where Black can capture it is surrounded by White stones: ```text . . . . . . . . X O . . . X O a O . . . X O . . . . . . . . ``` Black plays `a` and captures the White stone in the middle: ```text . . . . . . . . X O . . . X . X O . . . X O . . . . . . . . ``` Look at Black's new stone. It has exactly one liberty, the point that was emptied, so it is now in atari. If White plays there, White captures the Black stone and the board is exactly the first diagram again. Black could then capture again, then White, and so on without end. *What just happened:* a position can come back to itself. Without a rule, the game would have no way to finish. ## The ko rule The rule in the Japanese Rules of Go is direct: after you capture a stone in a ko, your opponent may not recapture on the very next move. In the position below, White may not play at `b`, the point where the stone was captured. ```text . . . . . . . . X O . . . X b X O . . . X O . . . . . . . . ``` White must play somewhere else first. After that, if Black has not protected the ko, White may recapture. That "somewhere else" move is the key idea. > 📝 **Terminology.** The move that gives you permission to recapture is played elsewhere on the board and forces a reply. It is called a **ko threat**. Capturing the ko stone back is called **retaking the ko**. ## Ko threats: how a ko is decided The rule says only "not immediately." Here is how a real ko plays out: 1. Black captures the ko. 2. White cannot retake yet, so White plays a ko threat: a move elsewhere that threatens something, such as capturing a Black stone or breaking into a Black area. 3. Black must decide. If Black ignores the threat and connects the ko (fills in the point so it cannot be retaken), Black wins the ko but lets White carry out the threat. If Black answers the threat, White is now allowed to retake the ko. 4. White retakes. Now Black may not retake immediately, so Black plays a threat of its own. The ko continues until one side runs out of useful threats or decides the ko is not worth any more. The player who has more threats of enough size wins it. The question to ask at step 3 is a trade: is what the threat takes bigger than what I lose by letting the opponent win the ko? If the ko decides a large group's life, you ignore small threats. If the ko is small, you answer almost any threat and move on. This is why experienced players say a ko is also about the rest of the board: it connects two parts of the game that have nothing to do with each other. > 💡 **Key point.** A ko fight is a trade of threats. You are not only fighting over one stone, you are spending your spare pressure on the board against your opponent's. ## Not every capture is a ko Ko needs both conditions: the capturing stone is left with exactly one liberty, and capturing it back would return the board to where it was. A capture where the recapture would not repeat the position is a normal exchange. A later phase shows the snapback, a sacrifice that looks a little like ko but is not (the sacrificed stones are recaptured immediately and the position never repeats). ## Beyond the basic ko The basic ko rule only stops an immediate recapture in one ko. Rare shapes can repeat over longer cycles, for example three kos at once (triple ko) or two kos where each player's capture keeps resetting the position. The Japanese rules handle these by allowing the game to be declared **no result** when the whole-board position repeats without progress, and the players replay it. Chinese-style rules and the American Go Association (AGA) rules instead ban any move that recreates an earlier whole-board position (a **superko** rule), so a repeating cycle cannot occur. Phase 4 compares the two approaches in the section on area scoring. For now, remember the part that matters: basic ko, "no immediate recapture," is the only ko rule you need to play on a 9x9 board. ## Check yourself ```quiz [ { "q": "After Black captures a stone in a ko, what is White not allowed to do?", "choices": ["Play anywhere on the board", "Recapture the ko on the very next move", "Pass"], "answer": 1, "explain": "The ko rule forbids an immediate recapture. White must play elsewhere first." }, { "q": "What is a ko threat?", "choices": ["A move that forces the opponent to answer elsewhere, so you may retake the ko", "A move that captures two stones", "A move that ends the ko by passing"], "answer": 0, "explain": "A ko threat is played away from the ko and threatens something. If the opponent answers it, you may retake the ko." }, { "q": "Why does the ko rule exist?", "choices": ["To make the game last longer", "Because without it the same position could repeat forever and the game could never end", "To give White an advantage"], "answer": 1, "explain": "Two players could capture the same stone back and forth without end. The rule breaks the loop." } ] ``` ## Recap 1. Ko is a shape where a capture can be answered by a recapture, repeating the position. 2. The ko rule: you may not recapture a ko on the very next move. 3. To retake, first play a ko threat elsewhere that your opponent must answer. 4. A ko is decided by who has more and bigger threats, and by how much the ko is worth. 5. Longer repeats are handled by "no result" in Japanese rules or by superko in Chinese-style rules. Next up, [Life and Death](03-life-and-death.md): how a group can be safe even when the opponent surrounds it completely. --- # Life and Death Surrounding a group does not capture it. A group can sit completely hemmed in on the board and still be impossible to remove. The reason is eyes, and understanding them is the biggest single step from "I know the rules" to "I can read a position." This phase explains why two eyes make a group permanent, which shapes make them, and the two traps (false eyes and seki) that fool every beginner. ## What an eye is, and why you need two An **eye** is an empty point (or a small pocket of empty points) enclosed by one player's stones, such that the opponent cannot safely play inside. To capture a group, the opponent must fill every one of its liberties. If a group has one eye, the opponent eventually fills all the outside liberties and then plays inside the eye, which takes the last liberty and captures. Here Black has one eye at the top, and every outside liberty is gone: ```text +------- | X . X O | X X X O | O O O . ``` White plays the empty point inside the group. The stone has no liberty of its own, but it takes Black's last liberty, so it captures all five Black stones (this is the capturing exception to suicide from Phase 1). Now give the group a second eye: ```text +------------- | . X . X O | X X X X O | O O O O O ``` Black has two eyes, at the corner point and the point between the two top stones. Each is a single point surrounded by Black. If White plays in either one, the White stone has no liberties and captures nothing (Black still has the other eye as a liberty), so the move is illegal. White can never fill both. The group can never be captured as long as Black does not fill in its own eyes. > 💡 **Key point.** **Two separate eyes make a group alive.** The opponent can only capture by filling all liberties, and filling either eye is suicide. A group with one eye is living on borrowed time. Terms you will see everywhere: a group with two eyes is **alive**. A group that cannot make two eyes is **dead**, even if it is still on the board, because the opponent can capture it whenever they choose. Strong players regularly stop playing around a dead group and use the saved moves elsewhere. ## The shape of the eye space You rarely start with two finished eyes. You start with some empty room inside your group, the **eye space**, and the question is whether it can be made into two eyes. The answer depends on its shape. These results come from playing out every possible sequence, with the group completely surrounded and no outside liberties: - **One or two points in a row: dead.** There is no room for two separate eyes. - **Three in a row: it depends on who moves first.** The middle point is the vital point (see below). - **Four in a row: alive,** whoever moves first. - **A 2x2 square of four points: dead.** It looks roomy but can only ever make one eye. Take three in a row: ```text +------------- | . . . X O | X X X X O | O O O O O ``` If Black plays the middle point of the three, two eyes appear on either side of it: ```text +------------- | . X . X O | X X X X O | O O O O O ``` But if White plays that middle point first, Black can no longer make two separate eyes. White's stone sits between the two empty points, and Black cannot make both safe: ```text +------------- | . O . X O | X X X X O | O O O O O ``` This is why the middle point is called the **vital point**: whoever plays it first decides whether the group lives. Finding it is a skill, but the idea is plain. Ask: "what single move leaves two separate eyes (or stops them)?" Here are the two four-point shapes. A straight line of four is alive; a square of four is not: ```text +--------------- | . . . . X O | X X X X X O | O O O O O O ``` ```text +--------- | . . X O | . . X O | X X X O | O O O O ``` In the square, White plays inside and the group can only ever end up with one eye. Try it on a board: it is the quickest way to feel why shape matters more than size. ## False eyes A **false eye** looks like an eye but is not, because the opponent can still reach it. The fastest test is to look at the diagonal points of an empty point. On the edge, an eye needs both of its diagonal points to be yours, and in the corner it needs its one diagonal. (In the open center the rule of thumb is that at most one of the four diagonals may belong to the opponent.) In this position the empty point on the top edge looks like Black's eye, but one of its diagonal points, in the second row, is White's: ```text +----------- | X . X X O | O X X O O | O O O O O ``` The corner Black stone has no liberty except that empty point, and the other four Black stones also have only that point. White plays it and captures everything, so Black has no eye here and this group is dead. The diagonal rule is a shortcut for the real test: can the opponent play there and survive? > ⚠️ **Gotcha.** Counting an eye that is not real is how beginners lose games they thought they had won. Before you relax, check the diagonals. ## Seki: mutual life One more outcome exists. Two groups can be stuck next to each other with neither able to capture the other. This is **seki** (Japanese for "mutual life"). Both groups share liberties, and whoever plays into those shared points first loses, so both players leave them alone: ```text +--------------- | O O O O X O | O . . X X O | X X X X O O | O O O O O O ``` The two empty points are shared liberties of the White group at the top left and the Black group around it. If Black plays one, that Black move reduces the Black group to a single liberty and White captures it. If White plays one, the same happens to White. So both leave them alone, and both groups live. In scoring, the points inside a seki count for nobody (Phase 4 covers counting). Seki is rare on a 9x9 board, but you should recognize it so you do not play a move that destroys your own group. ## Check yourself ```quiz [ { "q": "Why is a group with two separate eyes safe?", "choices": ["Because eyes make it bigger", "Because the opponent would have to fill both eyes, and playing in either one is suicide", "Because the opponent has to pass"], "answer": 1, "explain": "To capture, the opponent must fill every liberty. Each eye is a liberty, and filling one leaves the other, so the move is illegal." }, { "q": "Three eye points in a straight row inside a group. Who benefits from playing the middle point first?", "choices": ["Only the group's owner", "Only the attacker", "Whoever plays it first: the owner makes two eyes, the attacker kills the group"], "answer": 2, "explain": "The middle point is the vital point. The owner playing there leaves two separate eyes; the attacker playing there leaves the group with only one eye." }, { "q": "What is a false eye?", "choices": ["An empty point that looks like an eye but can be played by the opponent because a diagonal point is theirs", "An eye in the center of the board", "An eye with two points"], "answer": 0, "explain": "On the edge both diagonal points must be yours for a real eye. If one belongs to the opponent, the point is not safe." } ] ``` ## Recap 1. An eye is an enclosed empty point the opponent cannot safely play into. 2. Two separate eyes make a group alive, because filling either one is suicide. 3. Eye-space shape decides life: one or two points are dead, a 2x2 square is dead, a straight line of four is alive, and three in a row depends on the vital middle point. 4. A false eye has a diagonal point owned by the opponent; check diagonals before counting an eye. 5. In seki, two groups share liberties and neither can play there without losing, so both live and the shared points count for nobody. Next up, [Counting: Territory, Prisoners, and Komi](04-counting-territory-prisoners-and-komi.md): how a game ends and who wins. --- # Counting: Territory, Prisoners, and Komi Eventually both players agree there is nothing left to fight over, and the game has to be counted. Counting is where most beginners get stuck, mostly because Go has two common ways to do it and the two look different. This phase teaches the Japanese method on a finished 9x9 board, then shows the Chinese method on the same board so you can see they agree. ## How the game ends A game ends when both players pass **in a row**. Then comes the part computers do not need but humans do: 1. **Agree which stones are dead.** A dead stone is one that cannot avoid capture (Phase 3). Players remove stones they agree are dead. 2. **If you disagree, play on.** In the Japanese rules, a dispute about life and death is settled by resuming play. The player whose group is really dead will not be able to save it. 3. **Count.** You do not have to capture dead stones during play. That is why Go games do not end with endless capturing: a dead group is removed at the end, as if it had been captured. ## Japanese scoring: territory plus prisoners Three terms come up: - **Territory:** empty points surrounded by one player's living stones. - **Dame** (Japanese for "useless point"): empty points that touch both colors, so they belong to nobody and count for nothing. - **Prisoners:** stones you captured during the game, plus the opposing dead stones you remove at the end. **Score = territory + prisoners.** Stones sitting on the board do not count in Japanese scoring. Here is a finished game, simplified so the count is clear to follow. The White stone at row 3, column 2 is dead (inside Black's area) and the Black stone at row 5, column 8 is dead (inside White's): ```text 1 2 3 4 5 6 7 8 9 1 . . . . X O . . . 2 . . . . X O . . . 3 . O . . X O . . . 4 . . . . X O . . . 5 . . . . X O . X . 6 . . . . X O . . . 7 . . . X . O . . . 8 . . . X . O . . . 9 . . . X . O . . . ``` After the dead stones are removed, mark the points: `b` is Black territory, `w` is White territory, and `d` is dame: ```text 1 2 3 4 5 6 7 8 9 1 b b b b X O w w w 2 b b b b X O w w w 3 b b b b X O w w w 4 b b b b X O w w w 5 b b b b X O w w w 6 b b b b X O w w w 7 b b b X d O w w w 8 b b b X d O w w w 9 b b b X d O w w w ``` Count: Black has 33 points of territory, White has 27, and 3 points are dame. During the game, Black captured 1 White stone and White captured 2 Black stones. The dead White stone becomes one more Black prisoner and the dead Black stone becomes one more White prisoner. The point under a dead stone counts as territory (it is already included in the 33 and 27). | | Black | White | |---|---|---| | Territory | 33 | 27 | | Prisoners (captured in play + dead stones) | 1 + 1 = 2 | 2 + 1 = 3 | | Komi | 0 | 6.5 | | **Total** | **35** | **36.5** | White wins by 1.5 points. > ⚠️ **Gotcha.** Under Japanese scoring, playing a stone inside your own territory costs you a point (it covers a territory point and captures nothing). That is why good players stop playing inside their own area rather than "making it safe" for no reason. ## Komi: paying for the first move Black moves first, and the first move is worth something. To make the game fair, White gets a bonus called **komi** ("compensation") added to White's score. Modern Japanese professional games use **6.5 points** (the Nihon Ki-in, the Japanese Go association, moved from 5.5 to 6.5 in 2002). The half point also guarantees there are no draws, since scores are whole numbers. When playing an even game on a 9x9 board, use the same 6.5. If you play with a handicap (Black gets stones placed in advance), komi is normally reduced to almost nothing. ## Chinese area scoring: stones plus territory The other common method is **area scoring**, used under Chinese rules. You count the stones you have on the board, plus the empty points you surround. Prisoners do not matter. **Score = stones on the board + territory.** Komi is 7.5 in area-scoring games (the American Go Association, which uses area-style counting, also uses 7.5). On the same finished board, with the dead stones removed: | | Black | White | |---|---|---| | Stones on the board | 9 | 9 | | Territory | 33 | 27 | | Komi | 0 | 7.5 | | **Total** | **42** | **43.5** | White wins by 1.5 points again. Same winner, same margin. > 💡 **Key point.** The two methods count the same territory, but area scoring adds the stones on the board while territory scoring adds prisoners. A stone played inside your own area costs nothing under area scoring (the point still counts as yours) but costs a point under territory scoring. ### Why they almost always agree Compare the two scores for Black: area = stones on board + territory, territory = prisoners + territory. The stones on the board equal the stones Black played minus the stones Black lost (captured or dead), and the prisoners are the stones White lost. Work through the subtraction for both players and everything cancels except one term: **the margin of the area score exceeds the margin of the territory score by the difference in stones played.** Black plays first, so Black has usually played one more stone than White (12 against 11 in the example). That is exactly the one point the extra komi pays for: 7.5 instead of 6.5. So the two methods give the same winner and the same margin whenever Black has played one more stone than White. If the two sides have played equal numbers, the area score can come out one point different, which can flip the winner only in a game decided by half a point. A seki with eyes inside it is a rare special case where the two methods also differ. For ordinary games, learning one method teaches you the other. ### Superko variants Area-scoring rule sets handle repeated positions differently from the Japanese "no result". They ban any move that recreates an earlier whole-board position. Variants differ in detail: - **Positional superko:** you may not recreate any earlier board arrangement, regardless of whose turn it was (the Tromp-Taylor rules use this form). - **Situational superko:** you may not recreate an earlier arrangement with the same player to move (the AGA rules use this form). ## Check yourself ```quiz [ { "q": "Under Japanese territory scoring, which of these counts toward your score?", "choices": ["Your stones on the board plus your territory", "Your territory plus the prisoners you hold", "Only your prisoners"], "answer": 1, "explain": "Japanese scoring is territory plus prisoners. Stones left on the board do not count." }, { "q": "What are dame?", "choices": ["Empty points touching both colors that count for nobody", "Stones captured at the end", "The komi points"], "answer": 0, "explain": "Dame are neutral points. They are not territory for either player." }, { "q": "A Japanese-rules game ends with Black on 35 points and White on 30 points before komi. Komi is 6.5. Who wins, and by how much?", "choices": ["Black wins by 5", "White wins by 1.5", "White wins by 6.5", "The game is a draw"], "answer": 1, "explain": "White's total is 30 + 6.5 = 36.5. Black has 35. White wins by 1.5." } ] ``` ## Recap 1. The game ends when both players pass in a row; then dead stones are agreed (or play resumes if you disagree). 2. Japanese score = territory + prisoners; dame count for nobody; stones on the board do not count. 3. Komi compensates White for Black's first move: 6.5 in modern Japanese games, with the half point ruling out draws. 4. Area scoring counts stones on the board plus territory, with komi 7.5. 5. The methods agree because the difference between them equals the difference in stones played, which one extra komi point absorbs. 6. Area-scoring rule sets use a superko rule that bans repeating any earlier whole-board position. Next up, [Opening, Shape, and Tesuji](05-opening-shape-and-tesuji.md): where to play first, and how to win fights with tactics. --- # Opening, Shape, and Tesuji An empty board is paralyzing: 81 points on 9x9, 361 on 19x19, and every one of them is legal. Two ideas make the first moves less random: where stones are efficient, and a handful of tactics that win fights outright. This phase gives you the reasoning behind both, and every tactic here was played out move by move so you can trust the diagrams. ## Corners, then sides, then the center Territory is claimed by walls of stones. The edges of the board act as free walls, so the fewer open sides you need to wall off, the fewer stones you need. Take the same area of nine points (a 3x3 block) and count how many stones it takes to seal it off: | Where | Stones needed to seal 9 points | |---|---| | Corner | 6 | | Side (along an edge) | 9 | | Center | 12 | Here is the corner version. Six stones close off the 3x3 area, because the two edges do the rest: ```text +----------- | . . . X . . | . . . X . . | . . . X . . | X X X . . . | . . . . . . ``` That is the reason the standard advice is **corners first, then sides, then the center**: the same territory costs half as many stones in the corner as in the middle. Early in the game each stone you save is a tempo you spend elsewhere. This is a guideline, not a law: a stone in the center can influence the whole board, and strong players use it deliberately. As a beginner, take the corners and the area along the sides first, and keep your stones connected so they cannot be cut apart. On 9x9 the board is small enough that every stone is near an edge, so the priority becomes simpler: make one living, connected group, then use it to claim space. Do not scatter stones across the board; each separate stone has to be defended on its own. ## Tesuji: the clever move A **tesuji** (Japanese for "skillful play") is a tactical move that works in a local fight, the way a chess fork does. You find it by reading, which Phase 6 teaches. Four to know now. ### Double atari Atari on two groups at once. The opponent can only save one. ```text . . . . . . . . X X X X X . . X O a O X . . . . . . . . . . . . . . . ``` Each White stone has two liberties, and one of them is the shared point `a`. Black plays `a`: ```text . . . . . . . . X X X X X . . X O X O X . . . . . . . . . . . . . . . ``` Both White stones are now in atari. White saves the left one, and Black captures the right one: ```text . . . . . . . . X X X X X . . X O X . X . . . O . X . . . . . . . . . ``` *What just happened:* one move created two threats. The opponent has one move to answer, so one threat always succeeds. ### The ladder A **ladder** is a chase. A group in atari extends, which gives it two liberties, and the attacker puts it in atari again. This repeats along a diagonal staircase. White to play in this position: ```text 1 2 3 4 5 6 7 8 9 1 . . . . . . . . . 2 . . . . . . . . . 3 . . . O O . . . . 4 . . O X . . . . . 5 . . . . . . . . . 6 . . . . . . . . . 7 . . . . . . . . . 8 . . . . . . . . . 9 . . . . . . . . . ``` White ataris from below, Black extends, White ataris again, and so on. If every White move is atari, the chase runs diagonally to the edge. Here is the board after White has chased to the bottom right: ```text 1 2 3 4 5 6 7 8 9 1 . . . . . . . . . 2 . . . . . . . . . 3 . . . O O . . . . 4 . . O X X O . . . 5 . . . O X X O . . 6 . . . . O X X O . 7 . . . . . O X X O 8 . . . . . . O X X 9 . . . . . . . O . ``` The Black chain has one liberty, at row 9, column 9, and cannot take it: playing there would leave the chain with no liberties. White captures the whole chain next move. Reading a ladder is mechanical, and that is its danger: **check where it ends before you start it**. A ladder works for the chaser only if nothing helps the runner on the diagonal. ### The ladder breaker A **ladder breaker** is a stone already on the path. Add one Black stone at row 7, column 7 and run the same chase: ```text 1 2 3 4 5 6 7 8 9 1 . . . . . . . . . 2 . . . . . . . . . 3 . . . O O . . . . 4 . . O X X O . . . 5 . . . O X X O . . 6 . . . . O X X . . 7 . . . . . O X . . 8 . . . . . . . . . 9 . . . . . . . . . ``` The runner joins the breaker. The group now has seven stones and three liberties, so White's chase is over. White should not start this ladder. Every ladder question comes down to: "is there a stone of the runner's color near the diagonal path?" ### The snapback A **snapback** is a sacrifice. You let the opponent capture one stone, and in doing so they leave a bigger group with a single liberty, which you then capture. Study this corner. White's two stones have two liberties, at the corner and directly below it: ```text +--------- | . O X . | . O X . | X X . . | . . . . ``` Black throws a stone into the corner: ```text +--------- | X O X . | . O X . | X X . . | . . . . ``` The Black stone has one liberty, the point below it, and White can capture it there: ```text +--------- | . O X . | O O X . | X X . . | . . . . ``` But look at what White's capturing stone did: all three White stones now have a single liberty, the corner. Black plays the corner again, which captures all three: ```text +--------- | X . X . | . . X . | X X . . | . . . . ``` *What just happened:* Black gave up one stone and took three. If White declines to capture the thrown-in stone, Black captures the two White stones by playing below it instead. Either way White loses. > 💡 **Key point.** A snapback needs the capturing stone to end up in atari itself. If a sacrifice is not followed by an immediate recapture, it is a plain loss. Snapback and ko look alike (a capture followed by a recapture) but differ: in a snapback the recapture takes more stones, and the position does not repeat, so no ko rule applies. ## Your turn: shape and checking Check yourself before moving on: ```quiz [ { "q": "Why are corners usually played before the center in the opening?", "choices": ["The rules require it", "The edges act as free walls, so the same area takes fewer stones to claim", "Corner stones can never be captured"], "answer": 1, "explain": "Sealing a 3x3 area takes 6 stones in a corner, 9 on a side and 12 in the center, because the edges do part of the work." }, { "q": "What makes a double atari work?", "choices": ["The opponent cannot answer both threats with one move", "It captures two stones immediately", "It is the same as ko"], "answer": 0, "explain": "One move puts two groups in atari. The opponent saves one and you capture the other." }, { "q": "You are about to chase a ladder. What must you check first?", "choices": ["Whether any stone of the runner's color already sits on the diagonal path", "Whether it is move 10 yet", "How many prisoners you have"], "answer": 0, "explain": "A ladder breaker lets the runner connect and escape, which turns the chase into a loss for the chaser." } ] ``` ## Recap 1. The same area needs 6 stones in a corner, 9 on a side and 12 in the center, so corners come first, then sides. 2. A tesuji is a tactical move found by reading, not by pattern alone. 3. Double atari threatens two groups at once; the opponent can save only one. 4. A ladder is a diagonal chase of ataris; it works only if no stone of the runner's color is on the path (a ladder breaker). 5. A snapback sacrifices one stone so the capturer ends up in atari and you recapture a larger group. Next up, [Reading Ahead, and How AlphaGo Changed Go](06-reading-ahead-and-alphago.md): how to think several moves deep, and how a program learned to do it better than any human. --- # Reading Ahead, and How AlphaGo Changed Go Every tactic in the last phase came from the same habit: look at a position and play it out in your head before you touch a stone. That habit is called **reading** (in Go and in most strategy games), and it is the skill that separates players more than any other. This phase shows how to read, why the whole game is too large to read completely, and how a program learned to choose what to read, which is the story of AlphaGo. ## Reading: one branch at a time To read is to ask "if I play here, what does my opponent do, what do I do then" and keep going until the position is settled enough to judge. Three habits make it work: 1. **List the candidate moves.** Start with the forcing ones: ataris, captures, and moves that threaten something. 2. **For each, find the opponent's best reply,** not the reply you hope for. Assume they see what you see. 3. **Stop when the position is quiet** (no group in atari, nothing about to be captured), then judge it: who has more liberties, eyes, and space? Try it on a small problem. White's group in the corner is surrounded. Black to play and kill it. Three candidate moves are marked: ```text +----------- | a b O X | c O O X | O O X X | X X X X ``` White's group has an eye space of three points in a bent line: `a` is the corner, `b` and `c` are the two ends. Read each: - **Black plays `b` (or `c`).** The Black stone has only one liberty, `a`, so White captures it by playing there. That capture leaves White with two single-point eyes on either side of White's new stone, so White lives. - **Black plays `a`,** the point in the middle of the bent shape. Now White cannot make two separate eyes, and the group dies. *What just happened:* the answer is the vital point, the same idea as the middle of three in a row in Phase 3. A bent line of three behaves like a straight one: the point where the shape bends is the key. We confirmed `a` kills and `b` and `c` do not by playing out every White reply. ## Why you cannot read everything Reading a ladder is cheap because every move is forced. Reading an open position is not. In an average Go position there are roughly 250 legal moves, and a game lasts about 150 moves. Chess has roughly 35 legal moves in a typical position and games of about 80 moves (the figures are approximate and come from the AlphaGo research paper). Searching every sequence is out of the question in both games, and Go is far worse. For a long time, programs that tried to play Go by searching with human-written rules were far weaker than top professionals. Two problems stood in the way, and they are worth naming because they are the same two problems in every game-playing program: 1. **Breadth:** too many moves to consider at each turn. 2. **Evaluation:** deciding who is ahead in a position without playing it to the end. In chess you can count material. In Go, who is ahead depends on territory that does not exist yet. ## AlphaGo, 2016 AlphaGo, built by the company DeepMind, attacked both problems with neural networks (programs that learn patterns from examples) and combined them with a tree search. The system had three parts: - A **policy network** looks at a position and proposes the most promising moves, which cuts the breadth. It was first trained on about 30 million moves from games between strong human players. - A **value network** looks at a position and estimates who is winning, which solves the evaluation problem. - **Monte Carlo tree search** uses both to read ahead selectively, spending its effort on the most promising lines. After learning from human games, AlphaGo improved by playing large numbers of games against copies of itself. Its first public result came in October 2015, when it beat Fan Hui, the reigning three-time European champion and a professional, 5-0 on a full-size board with no handicap, the first time a program had done that against a professional. DeepMind's paper in *Nature* (Silver et al., "Mastering the game of Go with deep neural networks and tree search," January 2016) described it. In March 2016 in Seoul, South Korea, AlphaGo played a five-game match against Lee Sedol, one of the strongest professionals in the world. **AlphaGo won 4-1.** Lee won the fourth game with an unexpected move (the 78th) that the professional Gu Li called a "divine move." In the second game, AlphaGo's 37th move was an unconventional play that commentators called creative and unlike how humans play. All five games ended by resignation. A later version, AlphaGo Zero (2017), learned with no human games at all, only by playing itself, and beat the version that defeated Lee Sedol 100 games to 0. > 💡 **Key point.** AlphaGo did not read more moves than a human could in total. It chose which few moves to read and judged the result better than any earlier program, which is exactly what you practice when you narrow candidates and then evaluate a quiet position. You now know the whole stack in miniature: the rules, the tactics, the way a position is judged, and the search that ties them together. How machines in general play games (chess programs, minimax, Monte Carlo search) is the subject of [How Computers Play Games](/guides/how-computers-play-games). ## Where to go from here Play 9x9 games, and after each one find the one move where you misjudged a group's life or death. Replay it and read both sides. A 9x9 game takes about ten minutes, so a few dozen of them will teach you more than any book. Then move to 13x13 and 19x19, where the same ideas apply at larger scale. ## Check yourself ```quiz [ { "q": "In the corner problem, why does Black playing the middle point a kill, while playing an end point does not?", "choices": ["At the middle point White can no longer make two separate eyes, while an end-point stone is captured and leaves two eyes", "Because corners always die", "Because Black gets a ko"], "answer": 0, "explain": "The middle (vital) point stops two separate eyes from forming. A stone at an end is captured, which leaves White with two single eyes." }, { "q": "What was the result of AlphaGo's 2016 match against Lee Sedol?", "choices": ["5-0 for Lee Sedol", "4-1 for AlphaGo", "3-2 for AlphaGo", "A draw"], "answer": 1, "explain": "AlphaGo won four games to one. Lee Sedol won the fourth game." }, { "q": "Which two problems did AlphaGo's networks address?", "choices": ["Choosing a promising few moves out of hundreds, and evaluating who is winning without playing to the end", "Making the board smaller and shortening the game", "Rewriting the rules and the scoring"], "answer": 0, "explain": "The policy network narrows the choice of moves and the value network judges positions." } ] ``` ## Recap 1. Reading means playing a line out in your head: list candidates, find the opponent's best reply, and judge a quiet position. 2. The vital point of an eye shape (the middle of a bent or straight three) decides life and death. 3. Go has roughly 250 legal moves per position over about 150 moves, far too many to search completely. 4. AlphaGo combined a policy network, a value network, and Monte Carlo tree search, trained on human games and then by self-play. 5. It beat Lee Sedol 4-1 in Seoul in March 2016. 6. AlphaGo Zero later learned from self-play alone. Next up: [How Computers Play Games](/guides/how-computers-play-games) covers the search ideas behind programs that play chess, checkers, and Go.