Intelligent tutoring systems (ITS) have shown to be useful learning aids for helping students learn STEM subjects. Previous studies on ITS tend to focus on developmental aspects of the system, such as system design, programming architecture, and dialogue moves. In this systemic literature review, we focus on pedagogical aspects of ITS within STEM domains. Specifically, we identified the implemented scaffolding approach and the grounding on learning theories of ITS implementations. Specific research questions were: (1) what types of knowledge (i.e., conceptual learning, problem-solving, and model building) are delivered via an ITS within STEM domains? (2) what pedagogies or scaffolding methods are used to guide the ITS learning experiences? (3) what are the characteristics of the research designs and specific learning outcomes when learning with the ITS? The steps followed for performing this systematic literature review were: (1) identifying the scope and research questions, (2) defining the inclusion and exclusion search criteria of literature, and (3) classifying and cataloging the literature sources that use ITS for STEM in classroom research. The final data set is comprised of a total of 22 papers that meet our criteria. We found a lack of fine-grained research on the effectiveness of using ITS to improve the three major learning modes: conceptual learning, problem-solving, and model building, particularly in STEM domains. In addition, we recommend that research conducted on ITS and other learning technology aids should emphasize the utilization of well-established learning theories and pedagogical scaffolding methods so that ITS will be more accessible to STEM educators for introducing ITS to their students to better learn STEM subjects.

We developed a method to synthesize game levels that accounts for the degree of collaboration required by two players to finish a given game level. We first asked a game level designer to create playable game level chunks. Then, two artificial intelligence (AI) virtual agents driven by behavior trees played each game level chunk. We recorded the degree of collaboration required to accomplish each game level chunk by the AI virtual agents and used it to characterize each game level chunk. To synthesize a game level, we assigned to the total cost function cost terms that encode both the degree of collaboration and game level design decisions. Then, we used a Markov-chain Monte Carlo optimization method, called simulated annealing, to solve the total cost function and proposed a design for a game level. We synthesized three game levels (low, medium, and high degrees of collaboration game levels) to evaluate our implementation. We then recruited groups of participants to play the game levels to explore whether they would experience a certain degree of collaboration and validate whether the AI virtual agents provided sufficient data that described the collaborative behavior of players in each game level chunk. By collecting both in-game objective measurements and self-reported subjective ratings, we found that the three game levels indeed impacted the collaboration gameplay behavior of our participants. Moreover, by analyzing our collected data, we found moderate and strong correlations between the participants and the AI virtual agents. These results show that game developers can consider AI virtual agents as an alternative method for evaluating the degree of collaboration required to finish a game level.

Loot boxes are garnering increased attention in both the industry and media. One focal point of the discussion is whether loot boxes should be considered a form of gambling [1]. While parallels can be drawn between loot boxes and random reward schedules, researchers have argued that the “glorification” aspect of loot boxes that have heightened player awareness (e.g., opening a box, a pack of cards, or spinning a wheel) of randomness is a relatively new trend in games [2]. However, there is currently a dearth of empirical research on loot boxes. We make two contributions in this paper: 1) Infinite Loot Box, an open-source Unity platform for experimenting with loot boxes created from scratch, and 2) a 2×2 experiment (high/low visual effects x high/low audial effects; N=1235). We find that high audial effects significantly increase the number of loot boxes opened. Neither audial nor visual effects were found to significantly impact other variables. These contributions push forward our understanding of loot boxes and their contextual factors.

In this paper, we review the impact of generative artificial intelligence (AI) on game character creation. We critically examine the application of AI-driven computer graphics (AICG) technology across the character creation workflow, including concept generation, clothing design, modeling, props, cultural embedding, personality traits, and behaviors. Our research identifies potential applications, challenges, and future trends of these technologies in game development. Furthermore, it explores how AI and large language models (LLMs) can streamline workflows, automate asset generation, and reduce technical barriers in character creation. In this systematic review, we provide valuable insights for AI developers, game designers, and researchers.

The research community has long been interested in human interaction with embodied virtual characters in virtual reality (VR). At the same time, interaction with self-similar virtual characters, or virtual doppelgängers, has become a prominent topic in both VR and psychology due to the intriguing psychological effects these characters can have on people. However, studies on human interaction with self-similar virtual characters are still limited. To address this research gap, we designed and conducted a 2 (appearance: self-similar vs. non-self-similar appearance) × 2 (voice: self-similar vs. non-self-similar voice) within-group study (N = 25) to explore how combinations of appearance and voice factors influence participants' perception of virtual characters. During the study, we asked participants to collaborate with a virtual character in solving a VR jigsaw puzzle. After each experimental condition, we had participants complete a survey about their experiences with the virtual character. Our findings showed that 1) the virtual characters' self-similarity in appearance enhanced the sense of co-presence and perceived intelligence, but it also elicited higher eeriness; 2) the self-similar voices led to higher ratings on the characters' likability and believability; however, they also induced a more eerie sensation; and 3) we observed an interaction effect between appearance and voice factors for ratings on believability, where the virtual characters were considered more believable when their self-similarity in appearance matched that of their voices. This study provided valuable insights and comprehensive guidance for creating novel collaborative experiences with self-similar virtual characters in immersive environments.

‘Juicy’ or immediate abundant action feedback is widely held to make video games enjoyable and intrinsically motivating. Yet we do not know why it works: Which motives are mediating it? Which features afford it? In a pre-registered (n=1,699) online experiment, we tested three motives mapping prior practitioner discourse—effectance, competence, and curiosity—and connected design features. Using a dedicated action RPG and a 2x2+control design, we varied feedback amplification, success-dependence, and variability and recorded self-reported effectance, competence, curiosity, and enjoyment as well as free-choice playtime. Structural equation models show curiosity as the strongest enjoyment and only playtime predictor and support theorised competence pathways. Success dependence enhanced all motives, while amplification unexpectedly reduced them, possibly because the tested condition unintentionally impeded players’ sense of agency. Our study evidences uncertain success affording curiosity as an underappreciated moment-to-moment engagement driver, directly supports competence-related theory, and suggests that prior juicy game feel guidance ties to legible action-outcome bindings and graded success as preconditions of positive ‘low-level’ user experience.