Psychology
Using games to understand the mind
K. Allen, F. Brändle, et al.
This Perspective argues that progress in cognitive science, traditionally driven by simple, tightly controlled laboratory paradigms, can be expanded by leveraging games as experimental platforms. The authors pose the central question of how intuitive, enjoyable, and engineered game environments can help test whether theories developed in simplified settings generalize to more ecological contexts and enable new questions about inductive biases, intrinsic motivation, exploration, and complex multi-step planning. They set the context of widespread game play, recent growth in massive online games, and advances in statistical and computational modelling, motivating a systematic examination of the benefits, drawbacks, and recommendations for using games to study cognition.
The article synthesizes diverse prior work: classic lab tasks (two-step decision-making, n-back, multi-armed bandits, Towers of London, social coordination matrices) versus game-based tasks (4-in-a-row, procedurally generated video games, Sea Hero Quest, Little Alchemy 2, Virtual Tools, Overcooked-inspired cooking games, multi-agent construction). It reviews research on inductive biases (relational and object-oriented), decision-making in complex state/action spaces, and the role of objects in perception and planning. It discusses historical uses of games in psychology, gamification in education and therapy, and large-scale online data collections. Multi-agent games (Overcooked, Codenames) illustrate shared inductive biases and social phenomena across cultures. It compiles evidence on intrinsic motivation, curiosity, and play, and showcases large datasets (e.g., chess, Sea Hero Quest) enabling population-scale insights (navigation, diagnostics).
As a Perspective, the methodology is a structured synthesis and guidance rather than an empirical experiment. The authors: (1) define games for research purposes as intuitive, engineered environments that are intrinsically enjoyable; (2) analyze potentials of games (intuitive design revealing inductive biases; enjoyment enabling study of intrinsic motivation and large-scale, longitudinal data); (3) detail pitfalls (experimental confounds from game presentation, variability in prior experience, data access and infrastructure challenges, risks of ad hoc measures, modelling complexity, ecological validity constraints); (4) provide practical recommendations for using games: deciding between pre-existing versus self-made games; considering reward sparsity, complexity, curriculum, participant priors, and target populations; partnering with developers; recruitment strategies (standard platforms vs citizen science and app stores); data storage (databases), a priori predictions, effect sizes, conditioning on exposure, and including test levels; (5) propose analysis methods suited to complex game data (sampling-based log-likelihood estimation such as inverse binomial sampling; global optimization like Bayesian optimization); (6) advocate combined bottom-up (experiment to game) and top-down (game to controlled experiments) validation to reconcile internal and external validity.
- Games can reveal inductive biases in complex environments beyond traditional tasks. In the Virtual Tools game (3 × 600 × 600 action space), people compress action spaces via relational representations and can learn relational actions through limited trial-and-error.
- In Atari-like video games, object-based state representations are critical; replacing objects with texture patches undermines human play, highlighting the role of objects in constructing relational theories (e.g., keys and doors) that support efficient planning and exploration.
- Planning depth scales with expertise: in an online 4-in-a-row game, data from over 1.2 million players showed increased planning depth with expertise; online players began with worse search strategies than lab participants, indicating opportunities to study strategy improvement in broader populations.
- Multi-agent games (e.g., Overcooked) demonstrate the importance of shared inductive biases and iterative communication for coordination.
- Intrinsic motivation and enjoyment enable investigation of curiosity, exploration, and persistence without explicit rewards. Using Little Alchemy 2, researchers studied exploration in a goal-free setting and found empowerment-based strategies contribute to exploration.
- Citizen-science and enjoyable games can yield more conscientious behavior than monetary compensation alone (e.g., Skill Lab) and dramatically scale data and diversity: Sea Hero Quest collected navigation data from 4 million participants across 195 countries, informing national differences, environmental impacts on spatial skills, and personalized diagnostics for at-risk Alzheimer’s populations.
- Longitudinal gameplay enables tracking expertise acquisition and strategy change over extended periods; in Axon, initial performance differences predicted steeper learning curves with practice.
- Modelling complex game behaviors benefits from comparing classes of algorithms and identifying necessary features (e.g., tree search and feature dropping for sequential decision-making).
The Perspective demonstrates that games, by being intuitive and enjoyable, can extend psychological research into more naturalistic, complex settings while maintaining enough experimental structure to test cognitive theories. Findings illustrate how games reveal relational and object-based inductive biases, capture multi-step planning and social coordination, and unlock study of intrinsic motivation and play—domains hard to access in standard lab experiments. Large-scale, diverse, and longitudinal data from games allow robust tests of generalization and expertise development. The authors emphasize a complementary approach: use games to increase ecological validity and scale, but validate insights with controlled experiments or complementary games, and tailor computational models to the complexities of game environments. This combined strategy addresses the tension between internal and external validity and advances theory-building across cognition.
Games offer engineered, intuitive, and enjoyable environments that can substantially enhance research on the mind by improving ecological validity, enabling large-scale and longitudinal data collection, and opening access to questions about intrinsic motivation and holistic cognition. The authors recommend designing games around clear hypotheses and appropriate modelling frameworks, mitigating pitfalls through validation with controlled experiments and thoughtful data analysis. Future directions include leveraging games to study intrinsic goal-setting and persistence in sandbox contexts, integrating multiple cognitive systems within rich environments, examining cultural evolution of causal expectations in games, and advancing neuroscience and comparative cognition by collecting neural and behavioral data during gameplay across species. The synergistic use of games and classical experiments promises stronger, generalizable theories of human cognition.
- Reduced experimental control due to game presentation elements (‘bells and whistles’) can introduce confounds and limit generalizability.
- Greater variability in participants’ prior experience and inter-session activities can bias performance and yield non-representative data.
- Data access and infrastructure challenges: game companies may not share data; academic platforms often lack features for longitudinal tracking and progress.
- Risk of ad hoc, non-theory-driven measures in large datasets leading to spurious findings; need for predictive and concurrent validation of derived measures.
- Modelling complexity in ecologically rich tasks may require advanced machine learning/AI methods and comparison across classes of algorithms rather than single models.
- Games, while more ecologically valid than lab tasks, still do not perfectly capture natural behavior; ecological validity must be verified against real-world counterparts.
- Legal and financial risks (data protection, funding) can be more pronounced than for standard lab studies and require thorough planning and institutional guidance.
Related Publications
Explore these studies to deepen your understanding of the subject.

