Ethnomathematics, Decolonization, and Generative Artificial Intelligence: A Mixtape
Yuxi WEN [1], Yang YANG [2]
[1] Michigan State University, wenyuxi@msu.edu
[2] IOE - Faculty of Education and Society, University College London, dtnvany@ucl.ac.uk
Note: this manuscript has been accepted with minor revision for the 17th International Conference on Teaching Mathematics with Technology (ICTMT 17) at London, October 2025.
Abstract: Ethnomathematics has proven to be effective in igniting student interest as a form of culturally responsive pedagogy and decolonizing mathematics, but its effect coupled with multi-agent AI remain unexplored. This exploratory qualitative study developed first an ethnomathematics play, and then developed seven AI agents based on the characters of the play. Seven Chinese high school students participated in reading the play, debriefing the play, and then engaged with multi-agent AI. Results indicate that the play made mentally decolonized the partcipants to some extent, appealed to participant identity and improved participant agency. The multi-agent AIs went above and beyond than the play in that it not only reinforced decolonial thinking but appealed to general social-emotional learning and participant’s full identity rather than restricted only to ethnic and cultural perspectives, creating an “upside-down” relationship between human and AI.
Keywords: Ethnomathematics, Decolonization, Generative Artificial Intelligence
INTRODUCTION
Ethnomathematics: Definitions, Benefits, and Current Applications
Ethnomathematics, originally coined by Brazilian mathematician Ubiratan D’Ambrosio (1987), is defined as a “form of mathematics that articulates the relationship between mathematics and culture”, the core of which lies in the understanding that mathematical reasoning is deeply embedded within cultural practices (Ascher, 1998; Eglash, 1999). Ethnomathematics offers significant pedagogical advantages by making mathematics more culturally relevant, engaging, and empowering for learners, and is deeply decolonial in nature (Bernales & Powell, 2018). Empirical studies have shown that integrating students’ cultural practices—such as African designs, sports, or indigenous knowledge—can enhance mathematical self-efficacy, academic achievement, and cultural identity (Gerdes, 1999; Nasir & Hand, 2008; Näslund-Hadley et al., 2025). Furthermore, when adopted at the national level, as seen in China, Russia, and Japan, ethnomathematics supports culturally embedded learning at scale, fostering a deeper connection between students’ cultural heritage and mathematical understanding (Liu et al., 2025; Zhang & Shao, 2025).
Ethnomathematics is closely aligned with culturally responsive pedagogy (CRP), an asset-based instructional framework emphasizing the importance of connecting academic content to students’ cultural backgrounds and lived experiences (Moll et al., 2013; Rosa & Orey, 2011). Gay (2000) describes CRP as “using learners’ cultural knowledge, experiences, and perspectives as conduits for effective instruction,” while Ladson-Billings (1995) underscores its role in affirming cultural identity and fostering critical consciousness. Ethnomathematics reflects these principles by embedding mathematical ideas within historical and cultural narratives, offering students a more relevant and engaging learning experience, particularly for those who find traditional mathematics abstract or detached from their lived realities. However, the implementation of ethnomathematics faces several challenges. Curriculum resources that support the integration of ethnomathematics remain limited, especially in culturally heterogeneous classrooms where tailoring content to multiple traditions is a complex endeavor.
Multi-Agent AI in Ethnomathematics: Potential and Gaps
Multi-agent AI systems, comprising multiple intelligent agents with distinct instructional roles, offer new possibilities for supporting ethnomathematics-based instruction. In educational settings, these agents often take the form of pedagogical characters that simulate social learning dynamics (Córdova-Esparza, 2025). Empirical studies increasingly suggest that multi-agent systems can enhance both engagement and learning. For example, Nguyen (2023) found that 9th-grade students in science discussions benefited from both “expert tutor” and “peer” agents, with both leading to richer dialogues and deeper understanding. Cohn et al. (2025) proposed design principles for orchestrating teacher and peer agents in STEM learning, while Wang et al. (2025) reported improved programming outcomes and self-efficacy among undergraduates engaged in AI-supported collaborative learning. Zhang et al.’s (2024) systematic review of 44 studies found that while students often preferred certain agent roles or personalities, these preferences did not consistently correlate with learning gains, highlighting a need for further research into agent perception and learning outcomes.
Despite this growing body of research, the application of multi-agent AI to ethnomathematics remains underexplored. Current systems rarely feature agents capable of embodying distinct cultural epistemologies, storytelling traditions, or historical knowledge relevant to diverse mathematical heritages. There is also limited research on how agent behavior can dynamically adapt to the cultural identities of learners in real time. As a result, while multi-agent AI holds promise for advancing ethnomathematics, significant gaps remain in both design and implementation that warrant further interdisciplinary investigation.
METHODOLOGY
The research question, participants, artificats and its design principles, and sample and data collection method is described below.
Research Questions – This study focuses on exploring two research questions: (1) How does high school students’ epistemology of mathematics and its knowledge production change after exposure to an ethnomathematical and decolonial intervention supported by Multi-Agent AI? (2) How do Multi-Agent AI contribute to facilitating these epistemological shifts?
Artifact 1: A Play of Ethnomathematics – The author wrote a play – inspired by Imre Lakato’s Proofs and Refutations (1976) – but focuses more on the ethnomathematics and the math-making of the (so-called) Pythagorean Theorem. The two-act play is comprised of seven characters, six students and one Teacher, where the students are in fact personifications of how the Greek, Indian, Arabic, Italian, American, and Chinese culture produced a proof of the Theorem and views mathematical knowledge production in real history, as well as a Teacher figure that facilitates the conversation and kept the play going. The characters, along with the civilization, mathematical approach, and personality archetype that they represent, is included in the Appendix [1].
Artifact 2: Multi-Agent AI – The entire play was uploaded to ChatGPT (model type o4-mini-high) for initial analysis and summarization of the characters and their associated mathematical identities. The generated character profiles were reviewed and confirmed for accuracy. To further evaluate the capabilities of the multi-agent AI system, ChatGPT was prompted to assume different character roles and respond to questions representative of typical student inquiries or confusions related to mathematical concepts. Four example vignettes illustrating the interactions between the tuned ChatGPT agents and mathematical discourse are provided in the Appendix [2]. Following this systematic evaluation, the multi-agent AI system was deemed sufficiently developed for deployment in the study.
Participants – The study was conducted in late July 2025 in a rural town in Zunyi, Guizhou, China. Seven high school students (ages 16–19) were recruited through a convenient sampling and volunteering. All participants provided informed consent, with parental assent obtained when necessary.The sample size of seven was chosen to correspond with the number of characters in the play. For confidentiality, participants were anonymized as P1 – P7. Demographically, participants show a balanced representation across age, gender, and academic tracks, as well as diverse attitudes and proficiencies in mathematics, with details in the Appendix [3].
Process – The session lasted approximately 120 minutes and took place in a classroom setting. Participants and the researcher sat in a circle, each with a printed copy of the play. The session consisted of three parts: (1) Play Reading + Debriefing (40 + 40 minutes): Each participant was randomly assigned a character and read the play its entirety. Then, A semi-structured focus group interview was conducted, encouraging participants to reflect on the play, its key themes, and their perceptions of mathematics. The interview protocol is included in the Appendix [4]. (2) AI Interaction + Debriefing (20 + 20 minutes): Participants engaged with the multi-AI agent via the researcher’s laptop, taking turns to ask questions. Following this interaction, a brief debrief discussed the quality of AI responses.
Data Collection and Analysis – Qualitative data were collected through audio recordings of the focus group discussion and observations during the AI interaction. Participant responses and interactions with the AI were transcribed and thematically analyzed to identify shifts in epistemology regarding mathematics and the perceived impact of the multi-agent AI system. The analysis aimed to capture both individual reflections and group dynamics throughout the session.
RESULTS
The results section is organized around three themes in the debrief decolonizing experience, then moving toward discussing “mathematical taste”, and agency in learning mathematics. For each theme, the ethnomathematics part is first reported, in response to research question #1. Then, the effect of augmented multi-agent AI is reported, in response to research question #2.
Ethnomathematics as a Decolonizing Experience
The participants, by collectively debriefing on the epistemologies of mathematics of the characters, completed de facto a decolonizing experience on conceiving mathematical knowledge production. When the debrief got to Leonidas, everybody bursted into laughter when the name “Leonidas” was mentioned, as the play designed Leonidas to be painfully canonical on how mathematics is done, speaks like how people conceived school mathematics is, and brutally attacks anyone who disagrees with his epistemology of mathematics. P2, Leonidas’s actor, began the conversation:
P2 (laughs): Urghhh! It’s so exhausting (to read Leonidas’ lines)!
P5: Leonidas thinks mathematics is completely exclusive and built on “first principles”, and rejects everything that is outside of his system.
P3: I think Leonidas is basically autistic in doing mathematics. [Everybody laughs]
The debrief continued with participants commenting how Leonidas is often times rude, unforgiving, and indeed, sometimes almost autistic in his viewpoint. What the participants did not realize is that by being critical of Leonidas, they are in fact decolonizing from existing conceptions of what mathematics is and how mathematics is traditionally taught.
Participants are also witnessed how mathematics can be “personified” and how mathematics can be “liked” when personified. This is especially true for both P1 and P5, who are themselves girls on the Arts track [5]. P1 was randomly assigned to Valentia, also a female character with an artisitic approach to concpetalizing mathematics. P1 reflected (and strongly concurred by P5):
P1: I really think it was fate that I was assigned to read Valentia because I simply agree with her mathematical viewpoint. In fact, I don’t think Valentia cares about mathematics at all. She doesn’t really belong into this conversation [laugh]; she’s like an Arts student that stepped into a group of Sciences nerds by accident. She just wants to read Shakespeare and Dostovesky. I’m like her. In fact, I like her. That’s exactly how I felt like when I am learning mathematics.
The multi-agent AI portion was an opportunity for P1 and P5 to further decolonize herself from the “Leonidas”es and “Li”s. P1 and P5 are thoroughly attracted to Valentia as a play character and wanted to know Valentia more, as they strongly symthasized with Valentia, yet Valentia is someone who understands mathematics better than them, especially in the beauty of mathematics. During the AI session, they only talked to Valentia, and every question directed to Valentia was intended for Valentia to explain thoroughly what is beauty in mathematics, and how is mathematics “beautiful” in any case. In the AI debrief session, they admitted that they found themselves a mathematical role model. The play itself might have ended up only challenging their existing beliefs on the nature of mathematics and learning mathematics; but the AI part, for them, was an opportunity to immediately clear confusions, and experiment thoughts into realities, and find answers to questions.
An Opportunity of “Mathematical Taste”
Another decolonizing experience for the participants is that they discussed freely their “taste” on mathematical approaches although their mathematical skills varied greatly, showing that this topic is not reserved only for the “mathematical elites”. Participants were asked to identify which character most closely resembled their mathematics teachers they have had in their lives. Overall, there were two Sams, one Anish, one Leonidas, one Valentia, one Sayyad, and one who believes their teacher does not “deserve” any character, showing a great diversity in how participants perceived how their teachers thought about mathematics. Selected justifications are as follows:
P1: I honestly do not think my teacher deserve to be any character, because my teacher simply recites the official teacher’s manual. He has no personal interpretation of mathematics. I know this because I also bought a copy and his class are exact replicates of the manual, or at least 90% of it.
P2: Leonidas, because my teacher was very rude. He is like, “solve this question or get out of my classroom” type, so it’s very Leonidas.
P3: Sam, because Sam is actually quite smart. He’s like a guy who scavenges leftovers for a living, but he takes the leftovers from every house and makes a plate of fried rice out of it. I think that’s like my teacher on how to score high and answer questions fast on the University entrance examination (Gaokao).
P5: Anish, because to me mathematics is like a whim. My teacher would say some words I did not understand and then mathematics were made. It felt very “Anish” to me.
Most Chinese students can “taste” mathematics at most three times before entering university: their elementary, middle, and high school mathematics teachers. Sometimes, these three teachers would be more or less the same character. In the play, participants witnessed seven ways of thinking mathematics in a span of just 40 minutes, an incredible speed to build taste. Furthermore, in the multi-agent AI session, the participants are invited to talk to whoever it pleases the participant. The multi-agent AI thus augmented the play in that the play can only develop the mathematical taste of its participants, but faces difficulty to directly examine the taste after it has been developed. For example, it was not until the multi-agent AI portion that P3 admitted that he is a fanatic toward discovering mathematical symmetries and tesselations in everyday life. P3 then engaged in a series of questions towards AI-Valentia specifically on this topic. When asked why he would not talk to AI-Leonidas, P3 said that is might be afraid that AI-Leonidas would scold him. Therefore, the multi-agent AI not only satisfies participant’s taste of mathematics, it can also engage the participant in the exact topic of the participant’s choice rather than the textbook’s or even the play’s, therby giving even more authonomy to the participant in exploring mathematics.
Agency in Mathematical Learning
Another hallmark of decolonization is the theme of agency for the participants who believed they are incapable of doing mathematics or contributing to mathematical discussions, but were ignited when they discovered that they did in fact have much to contribute. This is especially evident for P4, who had struggled with mathematics after elementary school. As a personal friend of the author, P4 has always expressed her hatred, confusion, and learned helplessness in mathematics. However, she is highly engaged throughout the entire study and provided many valuable perspectives and ideas. Toward the end of the session, the author asked P4 if she was aware that she had just spent the past 90 minutes talking very in-depth about the nature of mathematics. Her face turned into disbelief. She shared, quite emotionally, that this is her first time doing mathematics for over an hour, and the first time that she had something to contribute in a mathematical discussion. This is echoed (again) by P1 and P5, who also self-identified as bad at mathematics. They felt that their presence in mathematics classrooms are “accidents” and they never have much to contribute. When the author asked P1 and P5 the same question (that they just spent 90 minutes talking about mathematics), they agreed immediately that talking this “would never happen for them” in their quotidian school contexts. When asked further that what changed in this session, they pondered, thought about it a bit, and said they couldn’t figure it out, but they “just had ideas to contribute” for this study/session.
Although P1, P4 and P5 could not elaborate why, it can be conjectured that cultural funds of knowledge, or the contexts and character developments given by the play has helped them to situate themselves in such a discussion. This is supported by P1 when she made this point earlier:
P1: I enjoy what I like and that is getting to know different types of people. This play is just about life as it is about mathematics. Therefore I’m less interested in what is being discussed, but how it is being discussed. Maybe I’m being too much of an Arts student, but even for another topic, these [six characters] might still have approached [discussing the issue] the same way.
Moving beyond ethnomathematics itself, multi-agent AIs also help create a nurturing environment where participants feel psychologically secure to engage and talk about their mathematics. All participants reported that their teachers operated on an ability-based model on mathematical competency, and low-ability students are not encouraged to contribute at all. For the play, the Teacher character served as a facilitator that encouraged all students to speak and never blamed students for not “getting” the mathematics. This is further exemplified for multi-agent AIs, which are designed to never get angry with users, are infinitely patient, and will attempt as many angles as possible to try to make the participant understand the point that the AI is illustrating, and praises user for every little piece of new understanding. In the AI session, all but one question is directed toward AI-Valentia. When asked why the participants want to talk to AI-Valentia so dearly (in addition to role modeling), participants said that AI-Valentia is incredibly patient and kind, and does not cherish her praise and acknowledging student growth. These are all important parts that keeps a learner interested in learning mathematics, but something that is so scarce in their lived educational experiences, and therefore the participants desired compensation in an AI environment. Perhaps it is the patience and praise for the multi-agent AIs that made P1, P4 and P5 to stay, because AI-Valentia made learning and discussing mathematics not just intellectually intriguing, but emotionally restorative experience.
DISCUSSION AND CONCLUSION
This study is based in ethnomathematics and culturally responsive pedagogy, but imagined a novel approach where multi-agent AI can augument the known power of existing decolonial pedagogies. The biggest result and contribution of this study is that its finding is somehow counterintuitive: this study created an “upside-down” world where the mathematical capacity of AI is limited, but AI can provide social-emotional value, and participants can find spiritual solace and role models in AI agents, but not their human mathematics teachers. For the ethnomathematics part (research question #1), participants had a first glimpse in experiencing decolonization of mathematics. However, the multi-agent AI experience furthered the processes as participants realize their misconceptions and mistreatments towards learning and understanding mathematics from a social-emotional perspective (research question #2). For the ethnomathematics part (research question #1), participants were motivated and appealed strictly in their ethnic and cultural identities; but with multi-agent AI, participants had agency to freely choose the part of identity that relates closest to them.
This study contributes to many existing gaps in literature. Results of this study show how multi-agent AI can activate student thinking and engage students in mathematical content. The most fundamental change of multi-agent AI against classical ethnomathematics is this: in traditional ethnomathematics, the facilitator would first appeal each participant to their cultural and ethnic part of their identity, and then present mathematics in relation to culture and ethnicity. Under this construct, the partipant must choose to be appealed through ethnic and cultural arguments, although identities are diverse aggregates. The participant had no agency in choosing which part of identity to be appealed with. However, with multi-agent AI, participants have full agency to discuss mathematics in the identity that they feel most comfortable and relatable with, as evident in the AI part of the session: participants freely chose the part that appealed to them most, and it does not necessarily have to be an ethnic argument anymore. As the field of AI education grows, the field is moving from theoretical frameworks to empirical research. This study contributes to the growing trend that calls for empirical evidence to feedback and advance theory.
The AI agent exhibited limitations in maintaining consistent character identities during extended interactions. Although identities could be restored through manual role-switching prompts, this process disrupted the interaction and introduced uncertainty. Almost all participants expressed that the AI agents went “out of character” in terms of mathematics after a few questions, and from a mathematical or philosophical point of view are not high-quality extensions of the original play, although it is incredibly helpful from a social-emotional point of view. Overall, AI is known to cause bias and furthers Eurocentric conceptions if left attended. This study questions if it can “do good”, and in what way, with proper guidance. Future research can explore further asnwers to this interesting question.
NOTES
1 – 4. Appendix link: https://drive.google.com/drive/folders/1nDI4kVU9o1bHf5FC6dXkoBEj0Q9tIQJa?usp=sharing
- In Chinese high schools, students must choose between the Arts/Humanities track or the Natural Sciences track.
REFERENCES
Ascher, M. (1998). Ethnomathematics: A multicultural view of mathematical ideas (1 CRC Press reprint). Chapman & Hall/CRC.
Bernales, M., & Powell, A. B. (2018). Decolonizing ethnomathematics. Ensino Em Re–Vista, 25(3), 565–587.
Cohn, C., Fonteles, J. H., Snyder, C., Srivastava, N., T S, A., Campbell, D., Montenegro, J., & Biswas, (2025). Exploring the design of pedagogical agent roles in collaborative STEM+C
learning. In Proceedings of the 18th International Conference on Computer-Supported Collaborative Learning (CSCL 2025). https://doi.org/10.22318/cscl2025.380628
Córdova-Esparza, D.-M. (2025). AI-powered educational agents: Opportunities, innovations, and ethical challenges. Information, 16(6), 469. https://doi.org/10.3390/info16060469
D’Ambrosio, U. (1987). Reflections on ethnomathematics. International Study Group on Ethnomatheamtics Newsletter, 3(1), 3–5.
Eglash, R. (1999). African fractals: Modern computing and indigenous design. Rutgers University Press.
Ernest, P. (2024). The philosophy of mathematics education: State of the art. Philosophy of Mathematics Education Journal, (42), 1–26.
Gay, G. (2000). Culturally responsive teaching: Theory, research, and practice. Teachers College Press.
Gerdes, P. (1999). Geometry from Africa: Mathematical and educational explorations. Mathematical Association of America.
Ladson-Billings, G. (1995). Toward a theory of culturally relevant pedagogy. American Educational Research Journal, 32(3), 465–491.
Moll, L. C., Soto-Santiago, S. L., & Schwartz, L. (2013). Funds of knowledge in changing communities. In K. Hall, T. Cremin, B. Comber, & L. C. Moll (Eds.), International handbook of research on children’s literacy, learning, and culture (1st ed., pp. 172–183). Wiley. https://doi.org/10.1002/9781118323342.ch13
Nasir, N. S., & Hand, V. (2008). From the court to the classroom: Opportunities for engagement, learning, and identity in basketball and classroom mathematics. Journal of the Learning Sciences, 17(2), 143–179. https://doi.org/10.1080/10508400801986108
Näslund-Hadley, E., Hernández-Agramonte, J., Santos, H., Albertos, C., Grigera, A., Hobbs, C., & Álvarez, H. (2025). The effects of ethnomathematics education on student outcomes: The JADENKÄ program in the Ngäbe-Buglé comarca, Panama. International Journal of Bilingual Education and Bilingualism, 28(5), 579–595. https://doi.org/10.1080/13670050.2024.2446987
Nguyen, H. (2023). Role design considerations of conversational agents to facilitate discussion and systems thinking. Computers & Education, 192, 104661. https://doi.org/10.1016/j.compedu.2022.104661
Polya, G. (1957). How to solve it: A new aspect (2nd ed.). Princeton University Press.
Rosa, M., & Orey, D. C. (2011). Ethnomathematics: The cultural aspects of mathematics, 4(2).
Wang, H., Wang, C., Chen, Z., Liu, F., Bao, C., & Xu, X. (2025). Impact of AI-agent-supported collaborative learning on the learning outcomes of university programming courses. Education and Information Technologies. Advance online publication. https://doi.org/10.1007/s10639-025-13487-8
Zhang, S., Jaldi, C. D., Schroeder, N. L., López, A. A., Gladstone, J. R., & Heidig, S. (2024). Pedagogical agent design for K–12 education: A systematic review. Computers & Education, 223, 105165. https://doi.org/10.1016/j.compedu.2024.105165