Multimodal AIeXperience Lab

Kumar, J. A. (2026, August). The sound of reasoning: Acoustic profiling of explainable AI feedback in ChatGPT Voice Mode [Conference presentation]. Ai4 Applied AI Research Conference, Las Vegas, NV, United States.

Abstract
A recent pilot study reflects the type of research the lab will support. Dr. Kumar presented “The Sound of Reasoning: Acoustic Profiling of Explainable AI Feedback in ChatGPT Voice Mode” at the Applied AI Research Conference held as part of the AI4 Conference 2026, Las Vegas. Using iMotions Voice Analysis powered by audEERING, the pilot study found that explainable AI feedback was associated with changes in ChatGPT’s vocal valence, dominance and arousal, particularly especially during negative feedback.

Kumar, J. A. (2025, July). AI on AI: Can GenAI tools design and evaluate course outlines better than we think? [Conference paper]. 2025 MIT AI and Education Summit. https://hdl.handle.net/1721.1/163143

Abstract
Despite the increasing use of generative AI (GenAI) tools in education, little is known about their effectiveness in producing pedagogically sound instructional materials. Therefore, this study evaluated the performance of six GenAI tools as instructional designers in generating a unit or module outline for an undergraduate course, focusing on developing learning objectives based on Universal Design for Learning (UDL) principles and later evaluating each outcome. Six free versions of GenAI, ChatGPT, Claude, Copilot, Gemini, Meta AI, and Perplexity were then analyzed thematically, focusing on instructional strategies, UDL integration, accessibility, and rubric development, while also being evaluated using a standardized points-based rubric by each GenAI tool. Findings revealed that Perplexity, Claude, and Gemini consistently produced stronger, learner-centered outlines, while Copilot and Meta demonstrated weaker instructional coherence. ChatGPT demonstrated strong instructional coherence but showed limitations in depth of rubric and integration of accessibility. Additionally, common limitations included insufficient timeline structuring, limited integration of learning domains beyond cognition, and weak alignment to summative assessments.

Kumar, J. A. (2024). Facial animacy in anthropomorphised designs: Insights from leveraging self-report and facial expression analysis for multimedia learning. Computers & Education, 223, 105150. https://doi.org/10.1016/j.compedu.2024.105150

Abstract
Anthropomorphism is the act of attributing human-like characteristics to non-human objects and has played a key role in the field of emotional design in multimedia learning. Despite its significance, the relationship between animacy and anthropomorphism, particularly facial animacy, remains underexplored albeit its potential impact on learning engagement and emotional responses. Hence, this study aims to address this gap by examining the effects of facial animacy in anthropomorphised designs using a 3 × 2 design (none vs. static vs. animated) based on self-reported measure (SRM) and facial emotion recognition (FER) and how both measures are associated. The findings revealed discrepancies between both measures, with mostly moderate to weak correlations between hypothesised associations. SRM results indicated that face animacy decreased perceived boredom, while static and dynamic anthropomorphised designs increased curiosity. The FER results revealed notable similarities between designs without anthropomorphism and the static versions, highlighting that facial animacy led participants to express more joy and less neutral expressions. Additionally, neutral expressions were associated with lower enjoyment perception, while negative emotions, especially boredom, were linked to reduced attention.

Kumar, J. A., Ibrahim, N., McEvoy, D., & Sehsu, J. (2023). Anthropomorphised learning contents: Investigating learning outcomes, epistemic emotions and gaze behaviour. Education and Information Technologies, 28, 7877–7897. https://doi.org/10.1007/s10639-022-11504-8

Abstract
Anthropomorphism is defined as attributing human traits and emotions to non-human entities. In the field of emotional design in multimedia learning, anthropomorphising essential learning elements has been associated with promoting positive learning experiences. Although it has been widely used for educational purposes, there are still limitations when considering different contexts, learning variables, and non-invasive measurement. Therefore, in this study, we investigated how anthropomorphising affects and associates with learning based on three perspectives: learning outcomes, epistemic emotions, and gaze behaviour. The findings indicate that anthropomorphism did not directly affect learning achievement, perceived satisfaction, and effort or when moderated by the need for cognition. However, anthropomorphism reduced the effect of perceived negative epistemic emotions, namely Bored and Anxiety. Additionally, a comparative correlation analysis indicated that anthropomorphism significantly reduced the perception of negative epistemic emotions for learning achievement (Confused and Frustrated) and effort (Frustrated). The gaze behaviour analysis revealed that anthropomorphism only influenced the initial view and not the number of views or dwell time. However, dwell time reflected partiality towards anthropomorphised elements showing negative emotions. The results implicate design and research considerations for future studies.