Doyle Brunson retired from poker in 2018. And at 87-years-old, one can’t expect him to stay on top of every new-fangled thing in poker. But off the back of watching the Negreanu/Polk grudge match, he’s become curious in the way solvers are used to study the game.
Tweeting to the world, Texas Dolly asked “Would someone please tell me what exactly are “solvers”? […] I don’t know all the hype about what to study.”
It’s a timely question.
The distinction between RTAs, bots, solvers, HUD, and AIs are not observed with much strictness in poker’s public discourse. The fuzziness of the language led to a pause in the Negreanu/Polk game earlier this year. But luckily Jason Koon was on-hand with a neat Twitter-length explanation.
Koon tweeted: “Hey Doyle! A user inputs an abstraction (simplified betting and raising options) and the computer uses an algorithm to “solve” a poker situation. Basically, the AI plays itself billions of times until both “players” reach a point where they can no longer exploit the other.”
This then is a sort of trial and error hunt for the Nash equilibria.
Man vs. Machine
There was a time, recounted in Brunson’s Super/System 2 when to figure out whether an under-pair was preferable to ace-king suited in an all-in pre spot, you had to sit down and deal it out until you had a statistically significant number of boards. The answer Amarillo Slim and Doyle Brunson came to after a night of this was that the pair has a slight edge.
Now, a computer could run a Monte Carlo simulation on that in moments.
However, some of the replies to the thread showed a kind of veneration for the old ways that betrayed a fundamental failure to “get” solvers.
For example, Nathan tweeted: “It’s modern-day snake oil to make insecure humans believe there is an answer to all of their problems at the table and that there is some mythical poker lord that can give them the unbeatable answers.”
Even Doyle seemed skeptical of the power of GTO and solvers. “All the solver folks come down to the Bellagio after we kick covid19’s butt,” he tweeted. “We will show u how to “solve” our games.”
AlphaGo go home
It is interesting to compare poker’s reception of computer training tools, to the game of go.
Go is thousands of years old and the level of human sophistication in the game is immense. Over time, sequences of moves called “joseki” have been worked out for certain situations. These are sequences of moves that are the best for both players.
After Lee Sedol (the best in the world at the time) was beaten 4-1 by Google’s A.I. AlphaGo in 2016, many josekis began to change.
The A.I. had shown better results could be gained from new sequences. Top players immediately integrated the new joseki as if AlphaGo were a new Meijin. Anyone who lacked behind lost a stone overnight. Studying with computers has improved human go playing immensely.
Though not Lee Sedol’s game. He retired right after the match. The computer hadn’t just beaten him on the board, but mentally too.
For now, though, the reluctance to involve solvers in player’s training seems rife enough that the cabal of top players like Polk and Negreanu who do rely on these tools have an enormous edge.
Poker isn’t diminished by a silicon coach. As Koon puts it in another tweet, “Solvers are useful tools but aren’t always intuitive. You need to understand the way poker works to interpret the data they give you. Poker is immensely complex.”
When it comes down to it at the table, Polk still has to beat a human being. And if Twitter has its way, that person will be Doyle.