Valve's upcoming MOBA-hero shooter, Deadlock, recently overhauled its matchmaking system, employing an algorithm surprisingly sourced from ChatGPT. This revelation, shared by Valve engineer Fletcher Dunn on Twitter (X), highlights the evolving role of AI in game development.
ChatGPT's Role in Deadlock's Matchmaking Overhaul
Dunn's Twitter posts detail how ChatGPT suggested the Hungarian algorithm as a solution for Deadlock's matchmaking challenges. This followed significant player criticism of the previous MMR system, with complaints focusing on uneven team skill levels and inconsistent matchmaking quality. Players reported consistently facing highly skilled opponents while being paired with less experienced teammates.
The Deadlock team acknowledged these concerns, promising a complete matchmaking system rewrite. Dunn's use of ChatGPT accelerated this process, leading to the implementation of the Hungarian algorithm. He expressed his astonishment at ChatGPT's utility, even dedicating a dedicated browser tab to the AI chatbot. However, he also acknowledged a potential downside: the replacement of human interaction and collaboration in problem-solving.
This sentiment sparked debate online, with some expressing skepticism about AI replacing programmers. Others voiced continued dissatisfaction with Deadlock's matchmaking, attributing recent issues directly to the ChatGPT-assisted changes. Negative feedback ranged from frustration over perceived matchmaking imbalances to accusations of negligence.
Algorithms are fundamental to data sorting and organization. In gaming, they can optimize team composition based on various parameters. Dunn's query to ChatGPT specifically sought an algorithm prioritizing one side's preferences in a two-party (bipartite) matching scenario. The Hungarian algorithm, chosen by ChatGPT, aims to address this need within Deadlock's matchmaking.
Despite the controversy, Game8 remains optimistic about Deadlock's potential. For a more in-depth look at the game and its playtesting experience, please see the link below.
