AI evidence extraction

Exploring Multimodal Generative AI for Education through Co-design Workshops with Students

Prajish Prasad, Rishabh Balse, Dhwani Balchandani · 2025

AI-generated evidence extraction, verified across multiple analytical personas. Not a substitute for the peer-reviewed original.

7/10

Relevance

1/4

Quality (LMQS)

Evidence

Citations

12.07

FWCI

Top 10%

Impact

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.1145/3706598.3714146

Methodology & findings

Study design

Two co-design workshops: (1) Ideation Workshop with 366 total participants (185 and 181 students across two sessions) forming 71 groups, where students created future learning journey maps; (2) Application Design Workshop (3-day design hackathon) with 30 participants forming 8 groups, where students created journey maps, storyboards, prompt flow diagrams, and UI prototypes.

Primary method

Co-design with participatory design workshops; Design Science Research methodology; Design Hackathon format

Main result

Students identified diverse educational problems that MLLMs can address, with the predominant problem area being "learning and skill acquisition" (21 out of 79 problems). Student applications demonstrated that "multimodal content creation, adaptation, and personalization" was the most prevalent MLLM experience category, with students designing applications to "create content in different formats, based on the users' preferences, needs and learning gaps." Additionally, "students incorporated various types of control mediums, ranging from standard inputs in text, to multimedia content such as documents, audio and video, to more specialized data sources like eye-tracking, face emotion detection and speech emotion detection" in their educational intervention designs.

Research paradigm

Constructivist/interpretivist with participatory design orientation

Author conclusions

"This paper describes experiences and insights from co-design workshops which we conducted with students as they worked on future MLLM-integrated educational applications. Through these workshops, students explored problems that MLLMs can address, and designed MLLM-integrated educational applications that can address these problems." The authors conclude that "the workshop design and activities can be adopted by the HCI community to conduct similar workshops in other areas as well" and that "these findings and research directions can guide students, educators and other researchers to further explore the opportunities and challenges of effective integration of Generative AI into educational applications."

Risk of bias

Selection bias: Recruitment via email invitation may have attracted only students with interest in AI/design; Sample bias: Limited to one South Asian liberal arts university, homogeneous student population; Facilitator bias: Authors acted as facilitators and coders, potentially influencing student ideation and analysis; Example bias: Provided examples during workshops may have constrained creative thinking; Confirmation bias: Focus on possibilities rather than limitations of GenAI applications; Selection bias: Participants self-selected through email recruitment for the application design workshop; Example bias: Student examples provided during workshops may have restricted creativity and biased ideation; Confirmation bias: Researchers focused primarily on possibilities rather than limitations of GenAI; Lack of demographic diversity: Study conducted in a single South Asian university, limiting generalizability; Selection bias: Participants self-selected for the design hackathon workshop; different from ideation workshop participants; Example bias: Authors acknowledge that examples provided during workshops may have restricted student creativity; Facilitator bias: All authors acted as facilitators during design workshop, potentially influencing student designs; Geographic/cultural limitation: Study conducted in South Asian context only; Insufficient emphasis on limitations: Workshops focused on possibilities rather than limitations of GenAI

Limitations

"First, the application design workshop was conducted with 8 groups (31 students) in a South Asian country, and hence the findings cannot be generalized to a larger population." Additionally, "some of the examples provided to students during the workshops may have biased their creativity, restricting them to ideas similar to the examples," and "our workshops focused on the possibilities of GenAI applications, and not enough on the limitations, drawbacks and challenges of using GenAI."

Open questions raised

Limited prior emphasis on GenAI educational intervention co-design with students
Need for training and co-design workshops with educators for content creation
Examination of feasibility and effectiveness of GenAI-generated content from educators' perspectives
Accuracy of personalization in various domains and topics
Granularity of personalization (course vs. topic level)
How intelligent tutoring systems can inform GenAI models

Data: Not mentioned; No datasets explicitly mentioned as available. Supplementary materials referenced (journey map template, instruction sheets, design principles worksheet, prompt flow diagram guide) but no direct URLs provided.Code: Not mentioned; No code repositories mentionedExtracted from: pdfAgreement 45%

Explore related topics

AI Literacy Training Human-AI Collaboration Design Science