Non-player characters, or NPCs, in games are usually scripted entities. They follow predefined movement patterns, respond through fixed behaviours, and rarely adapt in ways that feel genuinely individual.
In our Acta Ludologica paper, “Smart NPCs with Personality in a Serious Game Using Machine Learning,” we asked a different question:
What if NPCs could embody different personality traits and behave accordingly?
The paper was published on December 19, 2024, making December 20 a suitable date for a blog post explaining the research and its connection to MindEscape and SeriousMind.actaludologica
Traditional NPCs are designed through rules created in advance. A developer decides how an NPC moves, which action it takes, and how it responds to a specific event.
This approach is useful, but it can produce repetitive and predictable behaviour.
Our research explores how deep reinforcement learning (DRL) can be used to create NPCs whose actions are influenced by personality profiles. Rather than giving every virtual character the same strategy, we train agents to behave differently according to the Big Five personality model, also known as OCEAN:
The result is a team of AI-controlled characters with different behavioural tendencies.
Personality becomes visible through choices.
An NPC with a high Openness profile may explore unfamiliar areas, try alternative solutions, and show greater curiosity. An NPC with a high Conscientiousness profile may plan carefully, organise tasks, and persist until a challenge is completed.
Other personality profiles may influence communication, cooperation, risk-taking, conflict management, and reactions to uncertainty.
The goal is not to make NPCs “human” in every respect. It is to create controllable behavioural differences that reflect personality-related tendencies inside a serious game.
The study places personality-aware NPCs inside a serious escape-room environment.
The virtual agents must solve puzzles, interact with the room, exchange information, and coordinate actions. Each agent’s behaviour is shaped by predefined reward functions associated with a specific OCEAN profile.
This creates teams in which individual agents may:
The team’s success depends not only on the ability of individual agents, but also on how their personality-driven behaviours interact.
The escape-room environment includes Raven-inspired intelligence tests and a custom communication system.
This allows the NPCs to participate in tasks that involve reasoning, information exchange, decision-making, and cooperation. Their behaviour can then be analysed through gameplay metrics such as performance, stability, task-completion time, and interaction patterns.
The research found that different personality-driven agents produced different playstyles and behavioural tendencies. These variations affected how teams approached tasks and how efficiently they completed them.yadda.icm+1
Personality-aware NPCs can serve as more than game characters.
They can function as research tools for generating synthetic behavioural data. Collecting large datasets from human players is difficult, especially when researchers need examples of specific personality profiles or team combinations.
DRL agents provide a controlled environment in which researchers can:
This does not eliminate the need for human data. Human participants remain essential for validating whether the simulated behaviours correspond to meaningful human patterns.
However, NPCs can make it easier to explore controlled scenarios before conducting larger human studies.
The research also contributes to the development of MindEscape, SeriousMind’s flagship serious game for AI-powered personality assessment.
MindEscape places human players in immersive 3D escape rooms where they explore, solve puzzles, manage time, and respond to pressure. Personality-aware NPCs provide virtual behavioural baselines against which different gameplay styles can be examined.
Researchers can compare human players with trait-driven virtual teams and ask:
These questions connect NPC research with SeriousMind’s broader play-to-reveal approach.
The study also demonstrates how machine learning can support both game design and assessment research.
For game designers, personality-aware NPCs can make environments more varied and dynamic. Players may encounter virtual teammates who explore, plan, communicate, and react differently.
For researchers, the same agents can generate synthetic data and standardised behavioural profiles.
For assessment developers, they can help investigate how personality-related tendencies appear in gameplay and how different metrics might be used in a serious-game environment.
This creates a bridge between:
The future of NPCs may involve characters that are not simply scripted, but shaped by goals, traits, relationships, and context.
In a serious game, this could make gameplay more engaging while also creating richer research environments. Virtual agents may become collaborators, teammates, opponents, or behavioural reference points.
At SeriousMind, this research supports our broader goal of using AI and serious games to understand human behaviour, personality, teamwork, and potential.
A smart NPC is not valuable merely because it is more realistic.
It is valuable when its behaviour helps us learn something.
© 2026 SeriousMind. All Rights Reserved. Privacy Policy