The main argument I would like to put forth through this resource is that GenAI in education is not merely a technological advancement that brings in innovation (as positive) and academic integrity challenges (as negative), but GenAI, when uncritically examined, can perpetuate a colonial gesture. To be critical about GenAI and even to be critical about critical GenAI, we need to dive into some of the less discussed topics to be sufficiently informed when deciding to adopt (or not) GenAI in (higher) education.

This resource situates itself within the domain of decolonial and critical literacies which does not just critique GenAI at the surface but aims to deeply interrogate its onto-epistemological grammar and underlying assumptions and processes in its 'content' production.

It attempts to support educators and students in critically examining the use of GenAI in education to be well-informed (including recognizing complicity in harm), be more response-able, and adopt a pedagogy of refusal to resist the uncritical adoption of GenAI (McGranahan, 2016), to maintain democratic freedom and the place of education as a public good (Delahunty, 2024).

As Alterator et al., (2025, p. 14) posits: “GenAI perpetuates colonial hegemony in the form of essentialised presentation of culture, misrepresentation of authorship or entirely fabricated authorship.”

Strait et al (2018) demonstrate that people more frequently objectify and stereotype robots racialized as Asian and Black, than they do of robots racialized as White, highlighting racism and sexism.

Far from being neutral tools, GenAI systems perpetuate colonial logics by reproducing whiteness, universalism, and a monoculture of the mind, effectively erasing diverse epistemologies and ways of knowing.

Mbah and Jarvis (2026) argue that GenAI even risks to create an epistemicide and call for a Decolonial AI Sovereignty (DAIS) model.

Zembylas (2021) explores the intersection of decolonial theory and artificial intelligence (AI) within the context of higher education. He argues that AI technologies, as currently developed and deployed, often reproduce colonial structures of power, privileging Eurocentric epistemologies while marginalizing other ways of knowing and being. This process of algorithmic coloniality poses ethical and pedagogical challenges for educators, researchers, and students.

Tacheva and Ramasubramanian (2023) examine the AI Empire as heteropatriarchy, racial capitalism, white supremacy, and coloniality. They posit that AI technologies like ChatGPT and DALL·E are not neutral tools but are embedded in and perpetuate harmful practices such as extractivism, automation, surveillance, and containment. 

For example Zou and Schiebinger (2018) exemplify this with various examples such as:

- Google Translate defaulting to male pronouns,

- facial recognition misinterpreting Asian faces,

- word embeddings associating European American names with positivity, and African American names with negativity.

De Meulder (2025) investigates the ableist assumptions and the systemic oppression of deaf people's language through their exclusion in (Gen)AI design.

Park (2024) examine AI-generated images and their harmful biases that reproduce and perpetuate Whiteness, and provides reflective questions to engage with within the context of Art Education (p. 44):

- "What do you see?

- What patterns do you identify?

- Who/what is present or dominant?

- Who/what is missing or misrepresented?

- How are these patterns tied to existing social norms or stereotypes in contemporary media culture?"

Furthermore, Bolukbasi et al. (2016) expose how word embeddings (vector representations of language used in machine learning) can reflect and even amplify societal gender biases. They demonstrate that embeddings trained on sources like Google News strongly associate certain professions and roles with specific genders (e.g., "receptionist" with "female"). 

Hoffman et al. (2024) argue that both representational harms (e.g., stereotyping people) and allocational harms (e.g., unequal distribution of resources and opportunities) are already underway and risk intensifying if covert bias isn’t urgently addressed in how we evaluate and train AI systems.

GenAI literacy and its 'literacy' are topics often positioned in relation to "a sense of urgency" -which is typical of the White supremacist culture (Okun, 2025)- promoting speed and blind adoption over reflection and purpose.

As such, GenAI is a tool that promotes self-objectification

and is deprived of relationality and human judgement (Corbin, Tai, and Flenady, 2025) that are at the core of educational encounters. Therefore, its (ab)use risks the disappearance of people's (teachers and students) ‘subjectness’. 

Flenady and Sparrow (2025) argues that framing GenAI as a “study buddy” or “tutor” obscures its role as a mechanism of surveillance and conformity rather than liberation or support.

Furthermore, the authors posit that requiring students to take full responsibility for AI-generated content reflects a neoliberal, consumer-based model of education and undermines the communal, dialogic nature of scholarly learning.

The affordances of GenAi (e.g., speed, automation, adaptive) hail it as an efficient tool to provide feedback to students, yet AI-generated feedback is missing the social and emotional aspects of feedback by stressing only on scaling personalized comments and timeliness (Nazaretsky et al., 2024).

Trust and mistrust at the core of discussion of GenAI in higher education with the assumptions that students will cheat, be irresponsible and therefore needs to be taught to use GenAI responsibly (while teachers do not have to think about this) and controlled/monitored… 

AI detectors, based on suspicion and mistrust, turns classrooms into surveillance spaces, and teach students to cheat better or even to write worse to avoid being flagged (Giray, Roe and Diesta Espiritu, 2026).

Luo (2025) reveals how power asymmetries and one-way transparency erode student trust. Through concept mapping and interviews, the study finds that students often feel surveilled, fearful of being wrongly accused of cheating, and hesitant to explore AI tools due to unclear policies and punitive communication from teachers. While students are required to be transparent about their AI use, teachers rarely reciprocate by explaining how AI-influenced work is assessed, reinforcing top-down control. The study highlights the need for two-way transparency, improved teacher AI literacy, supportive policy communication, and collaborative policy design to foster trust and empower students in this evolving educational landscape. Luo advocates a shift from functional to relational approaches in AI policy and pedagogy, emphasizing trust as essential for navigating the uncertainties of AI-mediated learning.

Furthermore, Jafry and Vorstermans (2024) highlight that some students do not want to use AI and do not want to develop a dependency towards it. This perspective is often overlooked since (Gen)AI is largely positioned as a 'hype tool' that is inevitable to engage with in education.

The blind use of and uncritical engagement with GenAI in education reflect aspects of an Orwelian society by positioning students as objects of surveillance and control, not as autonomous subjects, with little real capacity to consent or opt out. Even when institutions claim transparency, the structural coercion of educational settings makes refusal nearly impossible, therefore manufacturing consent. 

Epistemology & Knowledge

• What kinds of knowledge, histories, and ways of knowing are being excluded in our use or teaching of GenAI?

• How does our institution reinforce or challenge the myth of GenAI's neutrality?

Neo-Coloniality & Imperialism in GenAI

Based on Nyaaba, Wright and Choi (2024):

• Whose knowledge is being centered, and whose is being marginalized by the data that train GenAI tools used in education?

  • How might these tools reproduce Eurocentric epistemologies and silence Indigenous, local, or non-Western ways of knowing?

• In what ways can educators actively resist and disrupt the neocolonial power dynamics embedded within GenAI systems?

  • How might teachers design pedagogical practices that use GenAI critically—not just as a technological aid, but as a site of analysis and contestation?

• To what extent does reliance on GenAI in educational research risk reinforcing colonial hierarchies of knowledge production?

  • How might scholars employ liberatory and human-centric design methods to reclaim autonomy over the narratives and data that shape educational futures?

• How can students be positioned not merely as users of GenAI, but as critical co-creators who interrogate and reshape its biases?

  • What pedagogical strategies enable learners—particularly from marginalized communities—to use GenAI as a lens for critiquing power, culture, and representation in digital knowledge systems?

Based on Jafry and Vorstermans (2024)

• How do AI tools store and use information?

• How are AI tools managed?

• Who benefits or is disadvantaged by AI tools?

• What are the challenges and opportunities in AI-based research and decision-making?

• How can AI be beneficial, particularly for students?

• Who is leading the discourse and development of AI, and which voices are missing or underrepresented?

(Un)critical AI (Il)literacy

The concept of AI lietracy is not neutral and often tends to focus on the equipment logics (technical skills such as prompting). While often claiming to also develop 'critical thinking', AI literacy paradoxically does not address criticality deeply and tends to maintain the status quo.

• How can students and educators recognize when GenAI produces false or misleading information, and what social or cultural assumptions shape what is presented as “truth”?
  How does this challenge our understanding of authority and knowledge in digital learning spaces?

• If GenAI outputs are never neutral, whose perspectives and values are being normalized through its design, and whose are being erased or assimilated?
  How might this process reinforce a monoculture of the mind where diverse epistemologies are suppressed?

• Who benefits economically and politically from the widespread adoption of GenAI tools in education?
  In what ways do market-driven interests influence what knowledge is prioritized, standardized, or commodified as “AI literacy”?

• What cultural narratives, histories, or identities are centered in GenAI-generated content, and what or who is only briefly featured, distorted, or omitted altogether?
  How can educators and students use these absences as a starting point for critical inquiry?

• To what extent does the illusion of GenAI “objectivity” mask its role in reproducing colonial or capitalist knowledge hierarchies?
  How can critical GenAI literacy empower learners to see through hallucinations not only as technical errors, but as reflections of systemic bias and epistemic exclusion?

• Does a lack of prompting skill or a refusal to use GenAI automatically render students or educators “AI illiterate”?
  What assumptions about technological participation, progress, and value are embedded in labeling certain groups as “illiterate,” and how might this reproduce deficit views of resistance or alternative epistemologies?

Agency, Power, Decision-Making

• How is consent obtained and maintained in learning environments that rely on data? 

• Are students and educators empowered to opt-out or question GenAI systems?

• Who is shaping GenAI (and GenAI literacy) curricula and policies? Do they reflect diverse lived experiences and epistemologies?

• How can we maintain students’ and teachers’ place as free subjects and not objects of other’s control? 

Data, Metrics, and Surveillance

• What is being measured, and why? Who benefits from these metrics?

• How might we resist or redesign data systems to reflect care, context, and justice?

• How is the GenAI data collected, processed and used? And are users (students, teachers) sufficiently informed about these aspects as well as the location of the data centres?

• How do we ensure relational trust? How do we avoid the slippery slope of punitive tones, policing students (e.g. random checks) and surveilling students’ (mis)use? 

Sustainability & Justice

• Are we critically engaging with the environmental and labor costs of GenAI?

• Are students and educators truly informed and able to critically examine the environment and social issues in depth surrounding AI use? (not just the obvious ones)

• What policies exist to ensure our GenAI choices do not exacerbate global inequalities?

• How are GenAI tools assessed in terms of their ethical conduct and contribution to harmful or violent practices? 

• How can we ensure that we are adequately informed about the environmental and social costs of our use of GenAI in education to make informed decisions of its use?

Hidden curriculum, Inclusion & Representation

• What underlying values, norms, and worldviews are reinforced through the implementation of GenAI and the ways AI literacy is taught?
  How might these practices reproduce dominant ideologies such as Whiteness, Eurocentrism, heteronormativity, or able-bodiedness under the guise of neutrality or innovation?

• Which ways of being, knowing, and expressing knowledge are marginalized or rendered invisible within GenAI-driven educational spaces?
  In what ways do claims of “objectivity” or “universal knowledge” obscure plural, embodied, Indigenous, or storied epistemologies that challenge Western ontological assumptions?

Assessment

• Are we only focusing on the production of learning outcomes and measurable outputs by students (following a performative logic) or are we engaging students in problem-posing tasks that connect to themselves and their (future) impact in the world?

• How can we ensure that we do not return to pen and paper of controlling and homogeneous/standardizing ways of measuring learning that are exclusionary?

• How might we rethink the ‘busy’ assessment tasks ‘given’ to students and move towards co-created assessments?

Zembylas (2021) calls for resisting digital neocolonialism and reshaping AI toward a more equitable future.

Adopt a pedagogy of refusal -as elaborated by Eggert, Hagen, and Turner (2025) and McGranahan (2016)- to resist simply following the crowd or the 'hype' and truly maintaining people's place a free subjects and not as objects of control.

Instead of homogeneity and 'one-size-fits-all' policy design that are rarely co-created with the people subject to it (i.e. students) and often accompanied with punitive tone, how might we 'pluriversalise' our policies?

Can we be more attentive to the underlying assumptions we make in these texts, and the power asymmetries that are created or maintained. For example, Luo (2025) demonstrates that in GenAI policies, students are the only ones who need to disclose the ways they use AI, but teachers do not have to be transparent, which erodes trust.

It is possible, as noted by Shahjahan (2016) to reimagine more equitable and pluralistic approaches to knowledge and policy-making in higher education.

Bolukbasi et al. (2016) advocate for the widespread use of debiased embeddings to prevent the reinforcement of stereotypes and call for hard and soft debiasing, distinguishing hard and soft debiasing:

• Hard debiasing removes gender associations from gender-neutral words while preserving necessary gender distinctions (e.g., maintaining the relation “she:grandmother :: he:grandfather”).

• Soft debiasing balances bias reduction with the preservation of original relationships for use in more nuanced applications.

Tacheva and Ramasubramanian (2023) reject the notion that AI can be reformed from within these oppressive systems, warning that superficial fixes only reinforce existing harm. Instead, they call for a radical, bottom-up transformation led by those most affected, highlighting recent collective resistance efforts, such as the unionization of African content moderators. Their conclusion insists that true justice and technological liberation can only be achieved through intersectional, transnational organizing that dismantles the foundational power structures of the AI Empire.

Repurpose: 

• Deliver learning activities with/without AI that reveal its fragilities, fabrications and breaking points

• Teach and actively support counter-hegemonic data projects: feminist, decolonial, minority-cultural, public and sustainable projects, projects of data sovereignty and data justice

• Encourage slow scholarship and slow learning with AI, refusing speed and productivity to focus on quality and interpretability;

Rebuild:

• Create and protect spaces for learning that do not involve data systems, such as hands-on making and investigating, live writing (not necessarily assessed), active experimentation, sensory feedback, discussing and debating, live (unrecorded) performance;

• Secure education policy and the training of teachers from the influence of AI lobbyists and AI-funded researchers;

• Redesign assessments away from instrumental outcomes and standardised rubrics; negotiate assessment tasks with students;

Refuse:

• Refuse to use systems or architectures that may be harmful, biased, unsustainable or unjust;

• Refuse to provide or collect data that may be used to build harmful data architectures or to make harmful discriminations concerning learners and educators;

• Refuse to adapt educational workflows and conditions of work to the use of AI, particularly where this will render education workers overworked, more precarious or less skilled;

Mirca Madianou

Centre for Countering Digital Hate (2025) conducted a large study that found that 53% of ChatGPT responses were harmful and 47% of these contained follow-up suggestions, some encouraging further engagement on harmful topics (self-harm and suicide, substance abuse, and eating disorders).

Because GenAI is so white, it creates a sociogeny (socially constructed phenomenon with biological effects) of racism that touch everyone and reproduces Fanon's (2008) 'white mask'. For example, it reinforces stereotypes, erases non dominant ways of knowing, and it creates anxieties, social isolation and dependence. It’s problematic for everyone to create homogeneous ‘knowledge’ based on an hegemonic worldview (White and Western) but particularly harmful for students of color or other minoritized groups, whose identities are not recognized, are devalued, and are erased.

A recent article at MIT by Chandra et al. (2026 - still needs to be peer-reviewed at the time I am writing this) showcases that sycophantic models like Chatbots cause delusional spiraling. It's not anecdotal, it's coded within the algorithm as and thus harmful for users, and particularly concerning for vulnerable groups like children.

GenAI detectors punish multilingual students the most (Giray, Roe and Diesta Espiritu, 2026).

While some researchers legitimize the use of AI for research, such as data analysis, a highly cited paper by Jowsey et al. (2025) completely rejects the use of AI in Reflexive Thematic Analysis (a dominant method in qualitative research).

Coining "eduslop" here, 🙋‍♀️! eduslop (noun) The amount of content (including, but not limited to, teaching materials, lesson plans, essays, homework, theses, exercises, guidelines, syllabi, feedback, assessments, etc.) that is "created" by using degenerative "AI" in educational contexts. Examples: - Robert used eduslop in one third of his course slides. - Anna shared her eduslop with other peers in class. eduslop (verb) The act of "creating" slop by using degenerative "AI" in educational contexts. Examples: - The teacher eduslopped his syllabus successfully. - John eduslops the essay minutes before the deadline is due, and doesn't need to invest time writing. Hallelujah! ___ Any other examples you want to add?!