Patches are a common feature in software, necessary for updates and problem fixes. In the world of esports, particularly in Multiplayer Online Battle Arenas (MOBAs), patches play a crucial role in altering gameplay dynamics. These updates can remove certain buildings, introduce new heroes, or transform existing ones dramatically. Subtler adjustments, such as a 5% reduction in ability cooldowns or item price alterations, also occur frequently. Understanding the impact of these patches on gameplay can be elusive. Typically, organizations like Bayes Esports monitor how professional teams adapt to these changing rules before insights can be drawn. This ongoing cycle ultimately leads to the emergence of a new “meta” until the next patch alters the landscape once more.
While this keeps the game engaging, it poses significant challenges for those setting betting odds. Side markets for MOBAs often revolve around objectives like defeating specific monsters, whose characteristics frequently change with each patch. Failing to update predictive models in light of these changes can expose betting operators to unforeseen risks from savvy bettors.
The task of updating these models, however, is time-consuming. New data must be collected to assess model accuracy post-patch, and understanding the potential shifts in gameplay—such as a change in the meta—takes time, often weeks. Despite expert analysts providing insights and human traders stepping in, human error remains a constant factor.
Now, imagine software capable of analyzing patch notes to predict shifts in the meta. This could revolutionize the industry. One approach could involve using self-play, where two teams of League of Legends or Dota 2 are simulated to experiment with new strategies until a new meta develops. By comparing shifts in strategic play and accumulating game statistics, new predictive models could be constructed much faster—transforming the model-recalibration process from weeks into mere hours.
This isn’t entirely improbable; OpenAI has demonstrated machine learning capabilities by teaching a computer to compete professionally in Dota 2. Modern AI has the potential to learn strategies, optimize team coordination, and work towards long-term objectives. The gap from analyzing gameplay patches to developing predictive models represents substantial progress that could soon be realized, especially with a dedicated team and quality data from sources like Bayes Esports. In a year, it’s conceivable we could rely more on AI assessments of patch notes than traditional expert analysis.
Dr. Darina Goldin, the director of data science at Bayes Esports, has a competitive gaming background, having played Team Fortress 2 during her graduate studies. While she no longer plays actively, her passion for esports remains strong. At Bayes, she has developed various predictive models for games like Counter-Strike, Dota 2, and League of Legends. Outside her analytical work, she engages in Brazilian Jiu-Jitsu training at the gym.
