I advocate for AI in education loudly, but I’ve never believed the tool is neutral. The same model that opens a door for one student can close one for another, depending entirely on how it was built and who it was built for. That tension is what Melissa Ozlem Grab (2026) puts under the microscope in her study of AI and culturally responsive teaching, published in Innovative Higher Education.
The Question Almost Nobody Asks About AI
Most research on AI in classrooms measures one thing: does it raise achievement? Grab points out that this framing leaves a hole. Nearly all the work on personalization focuses on cognitive skills and academic levels, with almost no attention to culture. Whether an AI tool honors a student’s cultural identity, surfaces voices from their background, or flattens everyone into the same dominant template is a separate question, and it’s the one she set out to explore.
Culturally responsive teaching treats students’ cultural identities and experiences as assets for learning, not obstacles to manage. Grab asks a fair question: can AI support that work, or does it get in the way?
She found that the same AI tools can be supportive and inhibitive to culturally responsive teaching at once. That rings true. It matches how the technology behaves in real classrooms far better than the breathless “AI personalizes everything” pitch I see in vendor decks.
What the Educators and Students Actually Said
Grab interviewed six educators and ran four focus groups with 22 students at a single institution, all of them already using AI-supported learning tools. The students came from diverse ethnic and cultural backgrounds, which was the point.
On the upside, educators valued AI for reading student data closely and tailoring instruction. A few students reported that their tools reflected their cultural identity and gave them a stronger sense of belonging. When the material connected to their culture, they said they were more motivated to learn. Grab also documents AI surfacing resources and voices from other cultures that students would never have found on their own, which several called a turning point in their whole experience with a class.
The skepticism was just as sharp. Educators and students worried that AI trained on unrepresentative data would reproduce and amplify the very biases culturally responsive teaching tries to dismantle. One student’s line cuts straight to it. Grab reports the student saying, “AI can give data and resources, but it’s not human. Sometimes, you just need a teacher who knows your background and experiences” (p. 13). That’s the whole debate in two sentences.
Bias, the Digital Divide, and Over-Reliance
Three worries ran through Grab’s data. The first was algorithmic bias. If the training data doesn’t represent a student’s culture, the tool’s recommendations won’t either, and the gap gets coded in as if it were neutral. This connects directly to what Warr and Heath (2025) describe as a hidden curriculum inside generative AI, the values and assumptions baked into a model that nobody voted on.
The second worry was the digital divide. Students noticed classmates without equal access to devices falling behind, and they called that unfair when AI was boosting everyone else. The third was over-reliance. Both teachers and students kept returning to the human relationship at the center of teaching, insisting AI should support it and never stand in for it. That echoes the concern Shaw and Nave (2026) raise about cognitive surrender, the slow handover of thinking to a machine that feels efficient right up until it isn’t.
Grab’s recommendations follow from all this. Build the tools with diverse stakeholders at the table, including students and community members. Train on culturally diverse data. Invest in real professional development so teachers know how to use AI in service of culturally responsive goals. None of that is surprising, but it’s correct.
Where the Study Falls Short
This is a single-institution study with six educators and 22 students, built entirely on self-reported perceptions. Grab is upfront about its qualitative scope, and that’s reasonable. But the findings mostly confirm what any thoughtful educator would predict: personalization helps, bias hurts, access is uneven, and the teacher still matters. The study maps attitudes well. It doesn’t show what AI actually does to culturally responsive practice over a semester, or whether students who feel more included also learn more.
That’s a starting point, not a fatal flaw. This intersection needs such groundwork before anyone can run the harder studies. I just want readers to weigh the evidence correctly. Perceptions are data. They aren’t outcomes.
What I agree with fully is Grab’s central design claim. She argues that “It is only by developing culturally responsive AI tools-intentionally including and seeking out diverse perspectives in the design process-that educators and developers can build AI systems that foster inclusivity, respect for diversity, and academic success for all students” (p. 9).
The fix lives upstream, in who designs the tool and what it’s trained on, not in a disclaimer bolted on after release. The same logic runs through Linsenmayer’s (2025) OECD work on AI and special education, where personalization either widens or closes gaps depending on intent.
Grab’s strongest contribution is a reframe. Culturally responsive AI isn’t a feature you toggle on. It’s a design commitment that starts before the first line of code and continues into how teachers are trained to use it. Her closing holds up: “any use of AI in higher education must be inclusive and thoughtful, not to break cultural sensitivity, but fair in the methodologies of teaching and learning” (p. 16).
References
- Grab, M. O. (2026). Teaching for equity: An exploration of AI’s role in culturally responsive teaching in higher education settings. Innovative Higher Education, 51(2), 859-880. https://doi.org/10.1007/s10755-025-09801-4
- Linsenmayer, E. (2025). Leveraging artificial intelligence to support students with special education needs (OECD Artificial Intelligence Papers No. 46). OECD Publishing. https://doi.org/10.1787/edu/wkp(2025)12
- Shaw, S. D., & Nave, G. (2026). Thinking fast, slow, and artificial: How AI is reshaping human reasoning and the rise of cognitive surrender. Working paper, The Wharton School, University of Pennsylvania. [https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6097646](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6097646)
- Warr, M., & Heath, M. K. (2025). Uncovering the hidden curriculum in generative AI: A reflective technology audit for teacher educators. Journal of Teacher Education, 76(3), 245-261.