Optimizing the human-AI cooperation: What can we learn from game design and in particular from NPC design?

Symbolic picture for the article. The link opens the image in a large view.

by Maximilian Wittmann

Artificial intelligence (AI) is generally regarded as a future key technology with tremendous potential to further optimize value creation in companies and to enable a wide range of improvements in everyday life. To fully tap this potential, fruitful cooperation between humans and AI is important. Successful human-AI cooperation plays a central role in various areas and application fields, for example in autonomous driving, human-robot interaction, or in supporting human decision-making with expert systems. However, cognitive (e.g. lack of comprehensibility of algorithmic decisions) and emotional challenges (e.g. prejudice, skepticism) often prevent effective cooperation between humans and AI systems. These challenges result in a lack of trust in AI-based systems, which in turn impacts technology adoption and use of AI. One potential solution for improving human-AI cooperation is gamification. Gamification describes the use of game elements in non-game contexts. The goal is to create game-like experiences (e.g., fun, satisfaction, motivation, immersion, feeling of flow) and to influence human behavior. 

A promising source of inspiration for embedding gamification elements into the human-AI cooperation is offered by the (video) game industry: In fact, player characters (abbreviated: PCs) seem to almost effortlessly build relationships with so-called non-player characters (abbreviated: NPCs) – characters in a game that cannot be controlled by the player) – trust is established, and the NPC-PC team works together toward a common goal. Consequently, future generations of AI systems could also benefit tremendously from considering the design patterns of non-player characters found in games. Therefore, these design patterns of NPCs need to be studied in greater detail. To overcome current barriers in human-AI interaction, we seek to understand why cooperation between humans and NPCs is so successful. Thus, the goal of our research is to generate structured knowledge about their frequency, effectiveness, and characteristics. Based on that, we can derive recommendations for more effective human-AI cooperation outside of games. However, recent NPC design research has primarily focused on studying design patterns of so-called “companions”, which represent a specific subcategory of non-player characters. Moreover, it currently remains an open question how NPC design elements from video games can be transferred to AI-based systems and which patterns might favor the success of cooperation between humans and AI.

In their study “What do games teach us about designing effective human-AI cooperation? – A systematic literature review and thematic synthesis on design patterns of non-player characters”, Maximilian Wittmann and Prof. Dr. Benedikt Morschheuser address this gap and explore design patterns of non-player characters. To answer the research question, a systematic literature review was conducted, followed by a thematic synthesis. The data analysis revealed the following interesting findings about the design patterns of non-player characters:

  • Six overarching design themes were identified that significantly influence human-non-player character cooperation: 1) responsiveness, 2) appearance, 3) communication, 4) emotions, 5) behavioral characteristics, 6) team structures.
  • Especially the NPC design patterns within the subcategories feedback and the ability to learn were used frequently and in many cases resulted in an improvement of the effectiveness of the cooperation in the sociotechnical system. 
  • 38 studies addressed or implemented NPC feedback mechanisms.

o NPCs are able to engage users through surprising or unpredictable twists and humor, and to induce behavior change through certain feedback mechanisms: Examples include open dialogues, presentation of ethical dilemmas, and embedding opportunities to reflect on past decisions made. Several studies applied decision milestones at which players were asked to evaluate their own decisions as well as those of the non-player characters. This confrontation served to critically assess what was happening and provide different perspectives.

o In certain contexts of use (e.g., real-time collaborative games with a PC and a physically present NPC), NPCs can stimulate human creativity. In particular, the combination of reciprocal (“turn-based”) game mechanics and intelligent robotic systems not only aroused participants’ curiosity, but also multiplied creativity and the quality of responses (e.g., in the drawing game called Magic Draw).

  • The subcategory ability to learn can be divided into three broad areas: 1) humans learn from NPCs, 2) NPCs learn from humans and become smarter based on human gameplay, and 3) NPCs learn from other NPCs.

o An effective methodology to evoke emotions in the player in a short period of time is to switch perspectives with the NPC. This swapping of points of views was investigated in 7 studies. Several authors staged NPCs in dramatic game situations, creating the impression of a vulnerable agent with weaknesses. Some studies also prevented the player from intervening in these scenes, such as an emergency situation involving trapped or burning robots. Often, the perspective shift was used as an introductory narrative, after which the player was allowed to enter the game world and take appropriate countermeasures to defuse the emergency situation.

o In two studies, AIs were trained the AI using human demonstration (“learning by demonstration”). Four studies used recording devices and external hardware (e.g., webcams, brain-computer interfaces …) to capture user states and algorithmically process the input data. Further processing and learning based on the input data often took place using advanced AI learning and optimization methods such as artificial neural networks and/or reinforcement learning. Authors used the obtained real-world data and player responses to trigger personalized and customized game events or adjust the difficulty level. This, in turn, can increase immersion and enjoyment of the game and the authenticity of the interaction.

o 3 studies demonstrated how NPCs learn from each other and how this can shorten learning cycles. In what is called “inter-domain learning” NPCs leverage the wealth of experience of other non-player characters. “Cross-domain learning” goes beyond this, as the skills of one domain are transferred to new domains. To illustrate, imagine an NPC agent who has learned to ride a bicycle. This knowledge could be used strategically in, for example, a new game environment or domain that requires the agent to ride a motorcycle.

The gained structured overview of effective design patterns that facilitate human-NPC cooperation is promising regarding the future design of AI systems outside of games. Software developers and designers of AI systems could particularly benefit from the following design principles: 

– Showcasing the vulnerability of the AI agent.

– Implementing the ability to switch perspectives with the AI agent and purposefully eliciting emotions from the user

– Explanation of decision making by diving into the “motivations” of the system and reflecting on an AI’s decision making on a regular basis.


This paper was accepted and published at the 6th International GamiFIN Conference 2022 (GamiFIN 2022).

What do games teach us about designing effective human-AI cooperation? – A systematic literature review and thematic synthesis on design patterns of non-player characters 

Abstract: Effective cooperation between humans and technologies powered by Artificial Intelligence (AI) is decisive to fully exploit AI’s economic and social potentials. However, the adoption of AI is often opposed by a lack of humans’ trust in AI systems and a dearth of interest in working with them. Turning to games for getting inspiration on how to optimize human-AI cooperation seems promising, since games engage humans almost effortlessly in interacting and cooperating with artificial non-player characters (NPCs). However, a structured overview on how game
design can optimize human-AI cooperation is missing in existing gamification  research. Therefore, this paper presents a systematic review of NPC design patterns and elaborates on what developers of AI systems can learn from game design. Guided by a thematic analysis, we present a structured overview of relevant design patterns clustered along six focus fields – namely I) NPC responsiveness, (II) appearance of NPCs, (III) NPC communication patterns, (IV) emotional aspects, (V) behavioral characteristics, and (VI) player-NPC and NPC-NPC team structures – which advance our understanding of designing and investigating cooperation
between humans and NPCs. The insights of  this paper can guide practitioners and future research regarding the design of more effective AI systems, the gamification  of human-AI cooperation, and the development of innovative NPC approaches.