In 2015, five poker players and one developer tried to build a winning poker bot. We had solver source code, ten servers, and more than a year of work. The bot never got past break even on the $1 games.
This is the story of my most fun failed project.
Costa Rica and the year of the solver
In 2015 I lived in Costa Rica with three poker friends. Poker was my whole focus. My game was improving, and I was moving up in stakes.
That year, poker solvers came on the scene. A solver is software that calculates the mathematically best play for a given spot. We got real nerdy about it. Every day we ran hands, studied the results, and had eye-opening moments.
Then the conversation took a turn. These solvers were so good that this might be the end of poker. Soon, bots would simply play like solvers.
Two of the four guys in that house were also software developers. So the next thought came naturally. Why not build one ourselves? A fifth friend, based in Thailand, wanted in too. Our idea was born.
Is a poker bot illegal?
You would think so. It wasn’t. A poker bot was a terms of service question, and the terms differed per site.
Some of the bigger sites in those days banned automated software. The smaller sites were different. Their terms clearly stated that automated software was allowed.
My guess is they valued volume and action over real humans. Maybe they also thought nobody could build a winning poker bot. About us, they turned out to be right.
Solver source code and ten servers
First we needed a brain. We contacted the maker of one of the solver programs and asked for source code access. After hearing our idea, he got excited too. He happily gave us special access.
Next came the hardware. We bought about ten servers. Server farms were an option, but the cost and locations made no sense.
Most of us were nomads. Only our friend in Thailand had a fixed home. So the servers moved into his house.
After Costa Rica, I flew there myself. I would literally sit in a room with ten servers running. It sounded like working inside an airplane, every single day.
Making a bot look human
A bot that plays well is only half the job. It also has to behave like a person.
For that, we hired an outside developer. He built the front end of the bot. It randomized the timing of every decision. Even the clicks were randomized, so the mouse never hit the same pixel twice.
That developer later joined me at AirdropAlert. A decade later, he still works here.
Why the bot never won
Here’s the funny part. We were a team of six. The players were winning regulars with good minds. We had special access to a solver, and we spent real time and money.
We still didn’t make a winning bot. It turned out to be much harder than we had mapped out. After a little over a year, we were barely break even on the $1 games.
So the experiment ended. Was it a waste of time? Sort of. Was it fun? A lot, actually. I was working on something I loved, from a brand new angle. There was also a potential shot at millions if we got it to win.
The ethics debate over dinner
Were there ethical debates? Sometimes. Every once in a while, over dinner, the same question came up. What if this actually wins?
Would we run it and pull millions out of the games? We all felt kind of icky about it, even though the terms allowed it.
Every option got a thorough hearing. Selling the bot once it won was one. Quitting the project was another. Running it ourselves was the third.
In the end, we never had to make the hard choice. The bot made it for us by not winning.
Are there poker bots today?
Do I think poker bots exist now? Absolutely.
We got to break even a decade ago, on the $1 games. That was a small team of friends with ten servers in a house.
Just imagine today. Solvers have gotten better. AI is now part of the picture. Servers and data centers are cheaper. The teams are probably more capable than we were.
I’m 100% convinced poker is solved well enough to beat humans at a very competitive level. My bet is that bots are running on every site.
What the bot taught me about automation
Looking back, this was an important step in my journey. I worked on a project in a team and created something new. Along the way, I learned how automation works.
That knowledge became important later, in airdrop farming. In my Poloniex lending story, the bots beat me. With the poker bot, I tried to be the one building them.
The logic is the same in both. At first, you can have an edge manually by being early. The next step is automation to scale it.
I still use that logic today. Sometimes a farm is profitable, and then I think of ways to scale it. It’s one of many lessons from poker that carried into crypto.
Final words
We never built a winning poker bot. What I got instead was a year of fun, a developer who stayed for a decade, and a lesson in scaling. Two years later, I quit poker and went all in on crypto.
For every story so far, browse the Morten crypto archives.
Credit: Source link
