BOOK 2 · PART I · PLAY · STUDENT CHAPTER
Read the Player, Set the Trap
Every opponent has habits, and robots have the most reliable habits of all. Learn to name the five kinds of player you'll meet, pick the trap that works on each one, and find out why a bot that never changes is a bot that people eventually beat.
Two players, one bot
When Jeff first heard about the AI agent, he texted back before he'd played a hand: "Bring on your AI Agent." He played it three times. The robot won the first and third; Jeff won the second. Then he went back to Las Vegas and finished 27th out of about 300 (Book 1, Part 17).
Jeff and I both learned to beat this bot, and we did it the same way: by watching it. Not its cards, its habits. Here are the two I use:
- It bets when you check to it. It reads a check as weakness. So with a strong hand, check, and let it put the chips in for you. That's exactly how my aces beat its kings in hand four of game 990544 (Chapter 3).
- When it checks a lot, it's weak. A bot that keeps checking usually has nothing worth betting. So bet the river, and it folds.
Neither trick is clever. Each one just uses a habit the bot can't see in itself and can't change. That's what this chapter teaches.
Five kinds of player
Poker players have sorted opponents into a few styles for a long time. Book 1, Part 11 used the same five when it planned how the AI could learn an opponent. Here they are, with how to spot each one and what to do.
| PROFILE | HOW YOU SPOT IT | WHAT WORKS AGAINST IT |
|---|---|---|
| Bluffer (a "Maniac") | Plays most hands. Raises a lot. Bets almost every time you check. Often shows a weak hand at the river. | Trap it. Check your strong hands and let it bet. Call it down with hands you'd normally fold. Never try to bluff it. |
| Calling station | Plays most hands, rarely raises, almost never folds. Reaches showdown a lot and loses there a lot. | No trap needed. When you have a good hand, bet it, and bet bigger. Never bluff: it will call. |
| Rock (a "Nit") | Folds most hands before the flop. When it does bet or raise, it has a real hand. | Steal from it. Raise its blinds often; it folds. When it bets big, believe it and fold. |
| Trapper | Checks a lot, then raises when you bet (a check-raise). Shows strong hands after playing quietly. | Don't walk in. Bet less often when it checks. Take free cards. Respect a check-raise. |
| Solid | Nothing stands out. Plays reasonable hands in reasonable ways. | No trick. Play your normal game. The mistake here is inventing a leak that isn't there. |
Real players are mixtures. Book 1, Part 16 read the AI agent in a 50-hand game as "tight, honest and readable": it folds a lot before the flop like a Rock, but almost never bluffs after the flop, so its bets there mean strength. Add the habit I found, betting when checked to, and you have a player with one Rock habit and one Bluffer habit. You don't need the perfect label. You need the habit.
Three traps, in poker words
- Slow-play. Act weak with a strong hand: check or just call instead of betting. It works against players who bet when you look weak, like the Bluffer. My hand four was a slow-play.
- Check-raise. Check, wait for the other player to bet, then raise. A slow-play with teeth. It works against the same players, and it's the Trapper's own favorite move.
- The steal. Bet or raise without a strong hand, because you expect the other player to fold. It works against the Rock, and against any player who has been checking a lot and is probably weak. My river bet against a checking bot is a steal.
A trap aimed at the wrong player wins nothing
Here's the part beginners miss. Every trap depends on the opponent doing something. A slow-play only pays if the other player bets. Try it against a Calling station and it just checks behind, the next card is free, and your strong hand wins a small pot, or loses to a card you gave away. Try a steal against a Calling station and it calls, every time.
So the order never changes: read the player first, then pick the trap. The hand journal now has a circle for it, "Setting a trap," so you can check later whether your read was right.
Why a bot that never changes gets beaten
Chapter 3 said robots are steady: hand one thousand is played exactly like hand one. That's their great strength, and it's also their great weakness. A habit that never changes is a habit people learn.
Jeff and I learned the AI agent's habits in a few games. The bot never noticed. It can't: every move it makes starts from nothing, with no memory of how the last hundred hands went. It can't learn from losing to us. People adapt. A bot that doesn't eventually gets read.
That's not a reason to give up on bots. It's a job description for the person in the loop: play the bot, find its habits, and write them down. Fixing them is the work of later chapters. In Part III, Jeff's list of habits becomes the instructions for a better bot. In the advanced exercises, you give a bot a memory so that it can learn an opponent too.
At the table
- Print a fresh hand journal. On the Sushi Menu, build a table of opponents you can read. Leave a few players blank: with no boxes ticked, they call every hand all the way to the end, just like a Calling station. Give a few others only the tightest opening row, Sklansky 1, so they should play like Rocks. Write down which seat is which.
Done when: the first hand is dealt and you know which seats should be Stations and which Rocks. - Play 20 to 30 hands. For each hand you write up, add who you were up against in "Something else, in my words."
Done when: you've written up at least five hands. - Try the right move on each. Against a Station: bet your good hands and never bluff. Against a Rock: raise its blinds, and fold when it bets big.
Done when: you've tried one move against each kind. - Sit down at the heads-up AI table. Try both of my traps: check a strong hand and let it bet; and when it has checked a lot, bet the river. Fill "Setting a trap" when you do.
Done when: you've tried each trap at least once. - Open both games in Game Review. For each trap: did the player do what you expected? Was your read right?
Done when: every "Setting a trap" circle has a Yes, No or Not sure beside it.
Try it yourself first
Four opponents. For each one, name the profile and the move that works.
- Seat 4 has called every bet for twenty hands. It has shown down five times and lost four.
- Seat 7 has folded before the flop on almost every hand. Now it raises from early position.
- Your opponent bets every single time you check, and twice turned over nothing at the river.
- Your opponent has checked three times this hand. You have a weak pair. It's the river.
ANSWER KEY
- Calling station. Never bluff it. When you have a good hand, bet more than usual; it will pay you.
- A Rock making a rare raise. Believe it. Fold everything but your best hands. Its habit of folding is your chance to steal its blinds on other hands, not this one.
- Bluffer. Slow-play your strong hands: check and let it bet. Call it down lighter than you normally would. This is the AI agent's habit when it's checked to.
- Probably weak, but read first. Three checks usually mean nothing worth betting, so a river bet often wins the pot. That's the steal. But if this player is a Trapper, the checks are bait: it's waiting to check-raise. One question decides it: what has this player done before when it checked?
Written by Kevin Swinson with Claudette, the pen name for Claude, the AI from Anthropic that helped build HoldemRobots.AI. The five profiles and their counters follow Book 1, Part 11; the read of the AI agent is from Book 1, Part 16 (game 990110); Jeff's matches are in Book 1, Part 17. The two habits Kevin exploits are his own account. On the Sushi Menu, a blank row plays every hand to the end (Kevin's description); how the other rows shape a robot's play is to be checked against the menu's explanation pages.