A group of researchers in Australia developed artificial intelligence (AI) models capable of accurately estimating the risk of gaming disorder in gamers. These models use data about the players’ relationships with their game avatars, their age, and the duration of their gaming involvement. The paper was published in Journal of Behavioral Addictions.
Since their emergence in the late 20th century, video games have become a fundamental part of modern popular culture. This rise coincided with the computer and internet revolutions, leading to a significant expansion in the gaming industry and the number of gamers. Today, video games are enjoyed by hundreds of millions of people worldwide.
Popular games like “Minecraft,” “GTA V,” and “Tetris” have each sold over 100 million copies. However, with the increasing popularity of gaming as a pastime and form of entertainment, health professionals have observed that some individuals engage in gaming in ways that adversely affect other aspects of their lives. This observation led to the coining of the term “gaming disorder.”
Gaming disorder is a mental health condition recognized by the World Health Organization (WHO). It is characterized by a pattern of gaming behavior that includes impaired control over gaming, prioritizing gaming over other activities, and the continuation or escalation of gaming despite negative consequences. For a diagnosis, this behavior pattern must be severe enough to significantly impair personal, family, social, educational, occupational, or other important areas of functioning. Generally, this pattern should be evident for at least 12 months for a diagnosis to be made, although this duration can be shortened if all diagnostic criteria are met and the symptoms are severe.
In their new study, Vasileios Stavropoulos and his colleagues aimed to investigate if AI and machine learning could predict the risk of gaming disorder in gamers. Prior research suggests that the nature of the bond between a gamer and their in-game avatar may indicate a risk of gaming disorder. Gamers who play games requiring deep engagement with their avatars, such as in role-playing games, might be at an especially high risk. With this in mind, the researchers set out to develop an AI model that predicts the risk of gaming disorder, based on factors like the user-avatar bond, immersion, the player’s age, and the duration of gaming involvement.
The study involved 565 role-playing gamers, with an average age of 29 years, ranging from 12 to 68 years. About half of the participants were male. They reported gaming experiences spanning up to 30 years, with an average duration of 5.6 years. Additionally, they had been using social media for an average of 7 years, spending about 3 hours per day on these platforms. Among them, 55% were employed full-time, 36% held an undergraduate degree, and 30% were single.
The authors assessed the participants twice, with a 6-month interval between assessments. By the second assessment, 276 participants had dropped out. The participants completed a gaming disorder diagnostic assessment (the Gaming Disorder Test, GDT-4) and an assessment of their bond with their in-game avatar (the User-Avatar-Bond Questionnaire, UAB-Q). The latter measured identification with the avatar (e.g., “Both me and my character are the same”), immersion (e.g., “Sometimes I think just about my character while not gaming”), and compensation (e.g., “I would rather be like my character”).
At the study’s outset, slightly less than 20% of participants were identified as being at risk of gaming disorder. The researchers then divided the dataset into two parts: 80% for training the AI models and 20% for testing their predictive quality. After training, the AI models could accurately identify participants at risk of gaming disorder based on their user-avatar bonding scale score, age, and gaming duration. They achieved this both with data from the study’s start and with data collected 6 months later. The level of avatar immersion was a critical factor in these predictions.
The study authors concluded: “The present study innovatively aimed to unlock the mental health diagnostic potential, likely embedded within the UAB [user-avatar bond], through the pioneering use of a sequence of different ML [machine learning] classifiers and emphasizing an individual’s disordered gaming risk. It did so while abiding with open science principles (i.e., accessible code and findings), such that research teams in the field can employ ML/AI [machine learning / artificial intelligence] to other already collected datasets related to the UAB to corroborate or negate the present findings. Furthermore, and in the context of the present study, ML/AI is converted from a game mechanic employed by industry to increase game engagement, and thus likely GD [gaming disorder] risk into a GD protective factor.”
The study explores a novel way to predict the risk of gaming disorder. However, the study relied solely on role-playing gamers and self-report data. It did not access medical records. Studies using other methods of data collection and involving players of other types of games might not produce equal results.
The paper, “Deep learning(s) in gaming disorder through the user-avatar bond: A longitudinal study using machine learning”, was authored by Vasileios Stavropoulos, Daniel Zarate, Maria Prokofieva, Noirin Van De Berg, Leila Karimi, Angela Gorman Alesi, Michaella Richards, Soula Bennet, and Mark D. Griffiths.