There’s a Catch-22 to competition design. When mastery comes too easily, the game quickly loses its appeal. Yet stiff competition entails high barriers to entry, leading to heavy churn at the beginner level. Ongoing research by Zhechao Yang, assistant professor of information systems and operations management at Costello College of Business at George Mason University, may have verified a solution: let AI play alongside humans (as long as the humans are none the wiser).
Yang’s working paper—co-authored by Qili Wang and Liangfei Qiu of University of Florida—looks at what happened after PUBG: Battlegrounds, a popular multiplayer battle royale online game launched in 2017, added AI-controlled opponents to the gameplay.
Game developers openly acknowledged that the lifelike bots, which PUBG incorporated in 2020, were designed to “improve the gaming experience for low-skilled players,” Yang says. “These players were frustrated because they struggled to eliminate more experienced opponents and were often eliminated early, so they didn’t have a good game experience and ultimately left the game.” The AI-controlled players essentially served as digital cannon fodder, matching the skill level of novice players and thus giving them more chances to score eliminations.
Yang and her co-authors accessed comprehensive gameplay data for 3,798 human players during the period December 2, 2019, to August 3, 2020. Because PUBG offers a team mode, in which players can form groups of four and compete with other teams, the researchers also tracked the formation and development of social connections within the game.
Across the entire sample, the researchers found that human players exposed to AI opponents participated in 16.7 percent more games on average and played for 49.3 percent more time each week, compared to players who were not exposed to AI opponents. Interestingly, the gains applied to all skill levels, not merely the novice players that the game developers primarily had in mind when they introduced AI.
“Intuitively, when you think about this policy, you would think it would harm the high-skilled players because the AI adds more low-skilled players to the game, presumably reducing the overall level of competitiveness,” says Yang. “But it’s actually the opposite. Because AI helps the novice players stay in the game long enough to improve, the low-skilled players move up in skill over time, and high-skilled players eventually have an expanded pool of capable opponents. It benefits the entire game ecosystem.”
Also, the addition of AI to the game increased play with friends, with players engaging in 8.8 percent more games with friends and spending 27.7 percent more time with them. Importantly, this social effect was not simply a byproduct of the increase in broader engagement; rather, it was a standalone statistical phenomenon indicating people were engaging differently, in addition to engaging more, with the game.
“Intuitively, when you think about this policy, you would think it would harm the high-skilled players because the AI adds more low-skilled players to the game, presumably reducing the overall level of competitiveness. But it’s actually the opposite. Because AI helps the novice players stay in the game long enough to improve, the low-skilled players move up in skill over time, and high-skilled players eventually have an expanded pool of capable opponents. It benefits the entire game ecosystem.”
— Zhechao Yang, professor in the information systems and operations management area
This social impact of AI was stronger for players who previously preferred randomly selected teammates over self-chosen partners (i.e., friends). After cutting their teeth on the virtual battlefield with the help of AI, these players became more active, supportive contributors to team gameplay, assisting teammates more often and becoming more willing to play with friends.
The researchers dug deeper to find out why. AI opponents made victories more attainable, but the effect was not simply about making the game easier. Because the AI opponents behaved enough like human players, victories over them did not feel trivial. Instead, players often interpreted those wins as signs that they were improving, which changed how they perceived their own ability and their role in the team. In turn, players became more willing to stay engaged and play with friends. As Yang describes it, “Novice players feel more capable and feel like they can make a contribution to the team, whereas before they might have felt their performance was too low.”
Yang stresses, however, that this shift depends in part on whether players perceive those victories as meaningful. “If people think the AI is obviously artificial,” Yang says, “they might think the AI is not so smart, which can change how they interpret the result and make the achievement feel less valuable.”
Yang believes this general effect could be harnessed outside video gaming. “The mechanism itself could be applied to learning any difficult skill or task, and it would be especially interesting for educational applications,” she says. “An AI-trained exercise with questions tailored to the beginning learner’s skill level might help build knowledge and confidence. Then the questions get harder and harder over time.”
Unlike most existing applications of AI, which aim to augment or substitute for human abilities, this approach suggests that AI can also shape behavior in a more indirect way: by helping people build confidence, persist longer, and become more willing to engage with others. Sometimes the best way to help people improve is not to remove difficulty, but to design early success in ways that feel earned—to let them, in a sense, fake it until they make it.