The video game market has grown rapidly in the decade and is now estimated to reach $143.5 billion dollars by 2020. The innovative nature of the industry has driven great advancements in software development, such as ultra-realistic 3D visualizations and physics modeling. But this dynamic industry has not seen much incorporation of modern machine learning and artificial intelligence applications into game development.
Industry leaders like Activision Blizzard (ATVI) are currently using machine learning and data mining on their business intelligence teams. Their goal is to better understand their player base to drive player engagement, which is increasingly key to a game’s success. Increased player engagement leads to more in-game purchases, brand loyalty, and product evangelism.
Blizzard has also joined several other game studios in making games and APIs available to notable AI research companies like Google’s DeepMind and Elon Musk’s OpenAI to help them advance the technology. Games such as Blizzard’s popular real-time strategy game, StarCraft II, are excellent training grounds for AI and machine learning research. Although researchers have seen success in developing AIs to defeat human players in well-known games like chess and Go, the enormous complexity of these competitive video games provide the next logical challenge. Yet, although the gaming industry is driving machine learning advancements in general, game companies still seem slow to apply the technology to their own game development.
This year, Activision rolled out “Call of Duty: WWII Skill” through a partnership with Amazon’s Alexa. In it, machine learning algorithms compare the way a player plays the game to that of all other players through cluster analysis, taking into account more and less successful playstyles and strategies. It then uses the Amazon Alexa speaker to deliver live coaching feedback and advice to that player while they play the game, making recommendations to hopefully improve success. Although the program is out of beta, there doesn’t seem to be much consensus on the effectiveness or usefulness of the technology. It is also a free application and Activision has said they currently have no plans to monetize it through an upfront fee or in-app purchases. Call of Duty: WWII Skill seems to be more of a proof of concept for the company, but it will hopefully lead to other new ideas .
The obvious evolution of machine learning in gaming is the application of machine learning techniques in the actual game development. One potential use being the creation of smarter, more realistic, and more responsive characters in game, that are able to learn from the actions of players and then react to and counter player tactics and strategies as well as develop the ability to deliver unscripted comments and responses when they observe player in-game actions.
Additionally, monetization models in gaming have been steadily shifting to lower upfront pricing (all the way down to free to play) balanced with increased in-game purchases and microtransactions which provide revenue over time. However, this model of course requires both long term player engagement and desirable in-game purchases. Machine learning can assist in the design of these products and in deciding when and how to present them to the player.
Another potential application for machine learning is in game community moderation. Gaming has struggle with cyberbullying and toxicity in in-game chat and player interaction for years. As gaming has become more mainstream and these unhealthy virtual interactions begin to have significant wider social implications, pressure to address this problem has increased. One recent study used machine learning to aid in data collection of cyberbullying activity from in-game chat logs. The machine learning algorithm was a sentiment text analysis service that attempted to identify and classify negative behavior in a popular online game. Although their implementation was somewhat ineffective and the researchers ultimately relied more on simple feature detection for bad language and racism, the idea is an interesting one. Creating a healthier, more inviting game environment for online players to dive into is almost certain to improve player engagement.
There remain a number of unanswered questions regarding the potential applications of machine learning in the gaming industry. First, improved adversary AI will likely lead to greater difficulty in games. Although this may be appealing to some players, others may find this to be more frustrating. A sense of individual improvement in a game increases player engagement, and playing against an unbeatable opponent may do just the opposite. Second, if machine learning models progress to the point that they are able to dynamically recognize player interests and desires and work with a “smart” game engine to procedurally generate and deliver that content, does this diminish the role of creatives in the game development process? These are just a couple of the secondary effects and factors to think about as this machine learning applications become more prevalent in the gaming industry. (798 words)
 Takahashi, Dean. “Newzoo: Game industry growing faster than expected, up 10/7% to $116 billion 2017.” Accessed at https://venturebeat.com/2017/11/28/newzoo-game-industry-growing-faster-than-expected-up-10-7-to-116-billion-2017/.
 Ewalds, Gaffney, and Vinyals. “DeepMind and Blizzard open StarCraft II as an AI research environment.” Accessed at https://deepmind.com/blog/deepmind-and-Blizzard-open-starcraft-ii-ai-research-environment/.
 Crecente, Brian. “Activision Bets on AI Video Game Coaching to Drive Success.” Accessed at https://variety.com/2018/gaming/news/ai-video-game-coach-alexa-1202764169/.
 Bertens, P., Guitart, A., Chen, P., & Periáñez, &. (2018). A Machine-Learning Item Recommendation System for Video Games.
 Murnion, S., Buchanan, W., Smales, A., & Russell, G. (2018). Machine learning and semantic analysis of in-game chat for cyberbullying. Computers & Security, 76, 197.