Game-Based Learning in the Age of AI

I learned to ride a bike by falling off it. Nobody gave me a manual. I tried, wobbled, adjusted and tried again. That cycle of doing something, seeing what happened and having another go is what a good game runs on, and it is why game-based learning (GBL) keeps turning up in classrooms and in workplace training.

Game-based learning and gamification

In GBL the game is the lesson, for example a trading simulation or a virtual patient. Gamification adds points, badges and leaderboards to a task that already exists. Hamari, Koivisto and Sarsa (2014) reviewed 24 empirical studies of gamification and found mostly positive effects, with large differences depending on context and on who was using the system.

What the evidence shows

Four large reviews give a fair picture. Here is how much research sits behind them.

129

papers reviewed by Connolly et al. (2012)

learning comparisons in Wouters et al. (2013)

comparisons in Sitzmann (2011)

media comparisons in Clark et al. (2016)

Wouters and colleagues compared serious games with conventional teaching. Games came out ahead on learning (d = 0.29) and retention (d = 0.36), but not on motivation, where the difference was not statistically significant. In effect-size terms these are small to medium gains: the average game learner scores roughly a third of a standard deviation above the comparison group. Gains were larger when the game was combined with other instruction, when players had several sessions, and when they played in groups.

Figure 1. Standardized mean differences (Cohen's d) reported in the abstract. Conventional benchmarks treat 0.2 as small and 0.5 as medium. The faded bar was not statistically significant.

Clark, Tanner-Smith and Killingsworth (2016) looked at digital games for K-16 students. Games beat non-game instruction by g = 0.33 across 57 comparisons. They also compared standard games with versions that had added design features, and those versions did better by a similar amount. Design matters about as much as the medium.

Figure 2. Clark et al. (2016). Both intervals sit clear of zero. They are also wide, so the true effect could be anywhere from small to roughly medium.

Sitzmann (2011) focused on workplace training with computer-based simulation games. Compared with other methods, trainees showed higher self-efficacy and more knowledge. Games worked better when trainees took an active role, when they could replay as often as they liked, and when the game sat inside a wider course. The same review found strong signs of publication bias, which means the true effects may be smaller than the published ones.

Figure 3. Sitzmann (2011): post-training differences for simulation-game groups relative to comparison groups. Bar lengths are proportional to the percentages.

Connolly and colleagues (2012) surveyed 129 papers on users aged 14 and over. Knowledge gains and emotional or motivational effects were the outcomes reported most often. The authors also point to methodological weaknesses in the studies they reviewed, so the headline numbers deserve some caution.

What AI could add

None of the studies above tested AI-driven games. What follows is my reasoning about where AI might help, and it has not been demonstrated by this research.

Figure 4. A conceptual model of an AI-adaptive learning game. Illustrative only; it is not drawn from a single study.

Adjustable difficulty

The game can raise or lower the challenge as you play.

Characters that answer

Language models let you negotiate or interview in free text.

Generated scenarios

New cases each round make answers harder to memorize.

Explained feedback

You hear why you were wrong while the decision is still fresh.

In finance, that could mean a portfolio simulation whose coach points out that you sold in a panic when the virtual market dipped. Healthcare, language learning, corporate training and cybersecurity have similar uses, such as virtual patients, conversation partners and phishing drills that get harder as you improve.

Risks worth taking seriously

AI can state wrong things with confidence. In finance or medicine a bad explanation does harm, so a person needs to review the content.

A game can be fun and teach little. Wouters et al. found no significant motivation advantage, which is a reminder that enjoyment and learning are separate outcomes.

Personalization collects data on how people think and where they slip. That matters most when the learners are children.

If the AI always supplies the hint, learners may skip the struggle that builds skill.

Most studies measure knowledge or confidence soon after play. Few follow behavior months later, such as whether people save more after a finance game.

A design checklist

Start from a clear learning goal and build the game around it. Pair the game with other instruction, offer several sessions, and consider group play, since those conditions raised effects in the research above. Keep a person in the loop for content review. Give feedback that explains reasoning. Measure outcomes beyond engagement.

Takeaway

The research supports games as a modest but real boost to learning, and the design of the game matters. AI may help with personalization and feedback, though that part is still a hypothesis. I would treat any claim of dramatic gains with suspicion, including the ones in vendor brochures.

References

Clark, D. B., Tanner-Smith, E. E., & Killingsworth, S. S. (2016). Digital games, design, and learning: A systematic review and meta-analysis. Review of Educational Research, 86(1), 79–122. doi:10.3102/0034654315582065
Connolly, T. M., Boyle, E. A., MacArthur, E., Hainey, T., & Boyle, J. M. (2012). A systematic literature review of empirical evidence on computer games and serious games. Computers & Education, 59(2), 661–686. doi:10.1016/j.compedu.2012.03.004
Hamari, J., Koivisto, J., & Sarsa, H. (2014). Does gamification work? A literature review of empirical studies on gamification. Proceedings of the 47th Hawaii International Conference on System Sciences, 3025–3034. doi:10.1109/HICSS.2014.377
Sitzmann, T. (2011). A meta-analytic examination of the instructional effectiveness of computer-based simulation games. Personnel Psychology, 64(2), 489–528. doi:10.1111/j.1744-6570.2011.01190.x
Wouters, P., van Nimwegen, C., van Oostendorp, H., & van der Spek, E. D. (2013). A meta-analysis of the cognitive and motivational effects of serious games. Journal of Educational Psychology, 105(2), 249–265. doi:10.1037/a0031311

Citation details and the numbers in the charts were checked on 4 October 2026 against abstracts and library records for each paper. The full texts were not re-read. Figure 4 and the section on AI are conceptual and are not backed by these studies.