BOOK 2 · PART I · PLAY · TEACHER CHAPTER
The Class Plays the AI
Day one, before anyone learns a rule. The whole class takes one seat against the AI agent, and every fold, call or raise has to be argued out loud. The AI writes down its reason for every move. By the end of the hour, so has the class.
One seat, thirty voices
A student playing alone makes a decision in two seconds and never says why. Put thirty students in one seat and that changes. Before the class can fold, call or raise, somebody has to convince the room, and that means saying a reason out loud.
Across the table, the AI agent does the same thing in writing. After every move it shows a one-line reason, like the one from hand four of game 990544: "overpair (Kings) on river, opponent checked, heads-up value bet with strong hand." So on day one, before anyone has learned a rule, the room gets to hear two kinds of reasons side by side: the class's, argued and voted on, and the machine's, written down every time.
This chapter is the first of two bookends. On the last day of the course, the class plays the AI again, this time with everything it has learned (Part III, "The Class Plays Again"). Keep what you write on the board today. You'll need it then.
What you need
| ITEM | NOTES |
|---|---|
| The heads-up AI table on a big screen | The table has its own login, separate from the demo account (Book 1, Part 7). Arrange access before class; Book 1, Part 25 says how to reach Kevin. |
| A whiteboard, with the hand journal's reasons written down one side | Before the flop: strong starting hand, good position, cheap to see a flop, weak cards, gut feeling, trying to bluff, something else. After the flop: I made a hand, drawing to something, the price was right, I think they're weak, they bet big and scared me, setting a trap, bluff, gut, something else. |
| 30 to 45 minutes | About 10 to 20 hands with discussion. One hand is enough to start a good conversation. |
| Three volunteers | A clicker who enters the class's moves; a scribe who keeps the board; a reader who reads the AI's reason aloud after each of its moves. |
How each decision runs
- Show the cards. The clicker reads out the class's two cards, the board and the pot.
Everyone can see what the seat sees, and nothing more. - Argue for a minute. Anyone can speak, but every speaker has to give a reason, not just a move. "Raise!" doesn't count. "Raise, because nobody has bet and our cards are strong" does.
At least two different reasons have been heard. - Vote. Fold, check, call or raise, by show of hands.
The clicker enters the move. - Write it down. The scribe marks the winning reason on the board, using the journal's words, with the hand number.
One tally mark beside one reason. - Hear the machine. When the AI moves, the reader reads its reason aloud. Ask the room: is that a good reason? Would we have said the same?
The AI's reason is said out loud, then the next decision begins.
Easy decisions don't need a debate. Seven-two offsuit facing a bet is a fold; the scribe just adds a mark to "easy folds." Save the room's energy for hands that are close.
What to listen for
- "I just have a feeling." Don't shut it down. Ask: what did the feeling notice? Sometimes it's a real read ("it always bets when we check"). Sometimes it's nothing. Either answer teaches something.
- Judging by the result. After a lost hand someone will say "that was a bad call." Ask: was it a bad call, or did we lose? Aces and kings both deserve to play for a big pot; one of them still loses.
- The room on tilt. After a painful loss, watch the next few votes. Classes get reckless just like people do. Point it out when it happens. It's the same lesson as Chapter 3, seen in thirty people at once.
- Someone spotting a habit. If a student notices that the AI bets whenever the class checks, or folds when the class keeps betting, celebrate it. That's reading a player, and it's the whole of Chapter 4.
After the game
- Count the board. How many votes rested on cards and price? How many on gut? How many on a read of the AI?
- Compare with the machine. Read three of the AI's reasons back. Which of the class's reasons could you imagine the AI writing? Which could it never write?
- Take a photo of the board. It comes back on the last day, when the class plays again.
Discussion questions
A student working alone can answer these on paper; they work just as well as questions to ask yourself.
- Which reason won the most votes? Was it the best reason, or the loudest?
- Did the class ever change its mind after hearing the AI's reason?
- Find a hand where the class lost but played well. Find one where it won but played badly.
- What did the room know about the AI by the end that it didn't know at the start?
NOTES FOR THE TEACHER
Question 1 is about group decisions, not poker. The loudest reason often wins. Good teams learn to ask for the reason behind a vote, which is exactly what a person in the loop does with an AI.
Question 3 repeats Chapter 3's main idea: judge the decision, not the result. If no hand fits, use Kevin's hand four: the AI's kings lost to aces, and the AI still played them well.
Question 4 is the start of Chapter 4. Any habit the class noticed, write it down. In Part III, a list like that becomes the instructions for a better bot.
Written by Kevin Swinson with Claudette, the pen name for Claude, the AI from Anthropic that helped build HoldemRobots.AI. The AI's reason quoted is from game 990544, hand 4. The table's separate login is described in Book 1, Part 7; Diversion 15 is in the diversions record of October 29, 2026.