The Art of Bluffing

Using Machine Learning to create better Poker Players

Machine Learning

Utilities

An American company wanted to explore whether AI could detect bluffing in poker to give their professional players an edge. We developed a model that identifies bluffing patterns, empowering players to spot bluffs and refine their own strategies.

Implementation time:

2-4 weeks

accuracy in detecting bluff

70%

professional players bluffs detected

2/2

ruined bluffs

+999

The Opportunity

The goal was to help players become better, as they dealt with their own human limitations:

Hidden signals

Subtle eye movements and decision-making cues were too complex to identify through observation alone

‍

Uncertain decisions

High-stakes choices were driven by instinct instead of reliable, data-backed insights

‍

Unclear patterns

Players couldn’t consistently recognize or learn bluffing behaviors across different games

‍

The Solution

As you may have noticed, this is not a usual project compared to the others done at DareData. However, it was an opportunity for us to put our computer vision experts to the test to solve such a unique problem.

‍

We provided our client with a model that could be applied during games to detect bluffing in their own players as well as their opponents. This allowed players to recognize opponents’ patterns and anticipate when they were bluffing.

‍

Our solution revealed that bluffing was closely tied to the eye behavior, including blink rate, fixation rate, and gaze direction at specific moments. Additionally, the time taken to make a decision also influenced whether a player was likely bluffing.

Meet the team of this project

Tiago Mota

Senior Machine Learning Engineer

João Rodrigues

Machine Learning Engineer

The Impact

With our model in their games, the players are now empowered by data to support their decision-making and make better decisions. What was once a choice based on gut feeling is now supported by analytics.

‍

From the players' perspective, they get an immersive playing experience, pushing their creativity to the limit as they find new ways to trick the model. Furthermore, the model can be used by players to learn their competitors’ signals.

‍

In fact, during the project, we successfully analyzed the bluffing behavior of two of the highest-stakes players. If you want to learn more, check out Tiago Mota’s blog post on his experience and challenges in this project.