Chapter 9: The Equity Imperative in AI Adoption
Who Gets Left Behind When Districts Adopt AI Without a Racial Equity Lens
Part III: HOW -- The Implementation
The superintendent was proud, and she had earned it. A $2 million AI tutoring platform, district-wide deployment, a press release that made the local news and the education technology blogs. Personalized learning for every student. Real-time feedback. Adaptive pathways. The vendor's slide deck had shown smiling children of every hue, and the pilot data from a suburban district in Connecticut had been impressive: gains in math fluency, improved essay organization, higher student satisfaction scores.
Six months after launch, her director of assessment brought the disaggregated data.
The platform was working beautifully for White and Asian students in the district's magnet programs. Their usage was high, their scores were climbing, their parents were posting about it on social media. But Black and Latino students in the same district -- students attending the same schools, sitting in the same buildings -- were having a different experience. The AI's reading recommendations skewed toward texts that did not reflect their linguistic or cultural contexts. The essay feedback tool flagged African American Vernacular English as "grammatical errors." The AI detection software the district had purchased alongside the tutoring platform was identifying Black and Brown students' original work as "AI-generated" at rates significantly higher than their White peers. And because the platform adapted to student performance, it was funneling struggling students -- disproportionately Black and Latino -- into repetitive drill sequences while routing high-performing students into creative, inquiry-based extensions.
The platform had not created these disparities. It had inherited them -- from the datasets on which it was trained, from the pedagogical assumptions embedded in its algorithms, from the educational system in which it was deployed. And then it had automated them, at scale, with the authority of artificial intelligence behind every recommendation.
The superintendent had asked the right question: How do we personalize learning for every student? She had not asked the essential question: Whose knowledge, whose language, whose experience is encoded in this tool -- and whose is erased?
This chapter is about learning to ask that essential question before a single AI tool enters a single classroom. It is about the equity lens that every other book on AI in education treats as an afterthought -- a chapter near the end, a paragraph on "digital equity," a nod toward "ensuring access for all learners." In this book, equity is not an afterthought. It is the architecture. It is the framework through which every adoption decision, every policy, every dollar spent on technology must be evaluated. Because if the Fourth Industrial Revolution is real -- and the data presented in previous chapters make clear that it is -- then the question is not whether AI will transform education. The question is whether that transformation will reproduce the same racial hierarchies that have structured American schooling since the Committee of Ten, or whether this generation of superintendents will have the moral courage and the practical tools to build something genuinely different.
No other book on AI in education makes this argument from the ground up. No other book connects the Fourth Industrial Revolution, superintendent leadership, and culturally responsive equity into a single, actionable framework. This chapter fills that gap -- not because I believe I have all the answers, but because I know what happens when the questions are never asked.
The Three Digital Divides, Revisited for the AI Era
When my dissertation examined the digital divide, the conversation was primarily about access -- who had devices, who had broadband, who could get online. That first divide has not been closed. It has been layered upon.
The Federal Communications Commission reported in 2024 that 24 million Americans still lacked fixed broadband access, with 25% of rural households going without compared to just 1.5% of urban households (FCC, 2024). The National Center for Education Statistics found that only 76% of rural students had fixed broadband at home, compared to 87% in suburban areas -- an 11-point gap that translates directly into an 11-point gap in AI readiness, because every AI-powered learning tool assumes a stable internet connection as its baseline (NCES, 2024). The State Educational Technology Directors Association documented that despite billions in federal E-Rate and Emergency Connectivity Fund investments, "sustaining progress to close the digital access divide" remains an active challenge, with funding cliffs threatening to reverse gains in the most vulnerable communities (SETDA, 2025).
This is the first digital divide: the access divide. It is the one policymakers recognize, the one that generates federal programs and philanthropic investments, the one that feels solvable because it is material. Buy more devices. Build more towers. Run more fiber. But access alone has never been sufficient, and in the AI era, it is less sufficient than ever.
The second digital divide is the skills divide, and here the data should alarm every superintendent reading this book. RAND Corporation's 2025 survey of American school districts found that roughly half of districts had provided AI training to teachers by fall 2024, double the proportion from the previous fall -- a rapid acceleration that suggests the field is taking AI seriously. But when the data are disaggregated by poverty level, a familiar pattern emerges: district leaders estimated that by the beginning of the 2025-2026 school year, nearly all low-poverty districts would have trained teachers on AI, while only six in ten high-poverty districts would have done so (RAND Corporation, 2025). The training divide maps onto the resource divide, which maps onto the racial and economic segregation of American schooling. Low-poverty districts -- which are disproportionately White and affluent -- are building institutional capacity for AI integration. High-poverty districts -- which serve disproportionately Black, Latino, and Indigenous students -- are not.
This is not a technology problem. It is a resource allocation problem with racial dimensions that Critical Race Theory demands we name. When Crenshaw (1989) established the framework for examining how race intersects with institutional structures to produce disparate outcomes, she was not writing about artificial intelligence. But she was writing about exactly this pattern: a system that distributes advantage and disadvantage along racial lines while maintaining the appearance of neutrality. A district that has not trained its teachers on AI is not "behind" in the way that a district with slower internet speeds is behind. It is structurally excluded from the capacity to make informed decisions about tools that are already in the hands of its students -- 86% of whom reported using AI during the 2024-2025 school year, regardless of whether their schools had policies, training, or infrastructure to support that use (Center for Democracy & Technology, 2025).
The third digital divide -- the outcomes divide -- is the most dangerous because it is the least visible. This is the divide between students who use AI tools that amplify their learning and students who use AI tools that reinforce their marginalization. It is the divide between personalization that responds to individual curiosity and personalization that replicates tracking. It is the divide between AI that sees a student and AI that sorts a student.
Digital Promise (2024) identified five key domains for equitable AI in education: Leadership for Digital Transformation, Coherent Systems and Resources, Consistent Access, Digital Competency, and Powerful Learning Propelled by Technology. What strikes me about this framework is that four of the five domains are organizational and leadership capacities -- not technology purchases. Equitable AI is not a product you buy. It is a culture you build. And building it requires confronting the reality that AI tools are not neutral instruments entering a level playing field. They are products of specific design choices, trained on specific data, deploying specific assumptions about what learning looks like, whose language is "correct," and whose knowledge counts.
Algorithmic Bias: How AI Reproduces and Amplifies Existing Inequity
In 2019, Ruha Benjamin published Race After Technology: Abolitionist Tools for the New Jim Code, and the title alone should have been a wake-up call for every education leader in America. Benjamin's central argument is that technology does not merely reflect existing racial hierarchies -- it encodes them, automates them, and gives them the appearance of objectivity. The "New Jim Code," as Benjamin defines it, is "the employment of new technologies that reflect and reproduce existing inequities but that are promoted and perceived as more objective or progressive than the discriminatory systems of a previous era" (Benjamin, 2019, p. 5).
This is not a theoretical concern for schools. It is an operational reality.
AI systems used in education are trained on datasets that overwhelmingly represent the Global North -- English-language, Western-normative, middle-class contexts that do not reflect the linguistic, cultural, or experiential realities of the majority of the world's students, let alone the majority of students in diverse American school districts (Frontiers in Computer Science, 2026). When an AI tutoring platform provides reading recommendations, those recommendations are shaped by whatever corpus of text the system was trained on. When an AI writing tool evaluates student essays, it evaluates them against patterns derived from that same corpus. When an AI detection tool determines whether a student's work is "original" or "AI-generated," it makes that determination based on statistical patterns that have been shown to flag non-native English speakers and speakers of non-dominant dialects at significantly higher rates (The Markup, 2025).
Joy Buolamwini's foundational work on facial recognition bias demonstrated that commercial AI systems had error rates of up to 34.7% for dark-skinned women compared to 0.8% for light-skinned men -- not because the technology was inherently racist, but because the training data underrepresented dark-skinned faces (Buolamwini & Gebru, 2018). The same principle applies to every AI system deployed in education. The algorithm is only as equitable as the data that trains it, the team that builds it, and the assumptions embedded in its design.
And here is where the workforce data becomes a equity issue: as of 2023, women made up less than 30% of global AI talent (OECD, 2025). The racial demographics are worse. The teams designing the AI tools that will shape how 50 million American public school students learn are overwhelmingly White, overwhelmingly male, and overwhelmingly from socioeconomic contexts that bear no resemblance to the communities those tools are meant to serve. As Safiya Noble (2018) demonstrated in Algorithms of Oppression, the absence of diverse perspectives in technology design is not a pipeline problem -- it is a power problem, one that produces tools that center the experiences, languages, and epistemologies of the people who build them while marginalizing everyone else.
Neil Selwyn (2022), one of the most important critical voices on AI in education, has argued that AI depends fundamentally on quantifiable data, but the most important aspects of learning -- social behavior, emotions, cognitive development, identity formation -- cannot be captured in numbers. When we reduce learning to data points that AI can process, we lose exactly the dimensions of education that matter most for historically marginalized students: belonging, cultural affirmation, the experience of being seen and valued for who you are rather than measured against a norm derived from someone else's experience. Selwyn cites scholars including Buolamwini, Noble, Benjamin, and Timnit Gebru in framing AI products as advancing "engineered inequality" in already inequitable social contexts -- a phrase that should be posted on the wall of every superintendent's office in America.
The OECD's 2024 report on AI's potential impact on equity and inclusion in education warned that algorithmic bias can lead to harmful decisions about school course schedules, grading, career counseling, and other outcomes that disproportionately affect marginalized students (OECD, 2024). These are not edge cases. These are core functions. When an AI system recommends that a Black student take remedial math instead of algebra based on a prediction derived from historical data that reflects decades of racialized tracking, the algorithm has not made an error. It has faithfully reproduced the logic of the system that generated its training data. The bias is not a bug. It is the data, operating as designed.
Trotter's Technology Extraction Applied to AI
In Chapter 1, I introduced Joe Trotter's (2000) concept of technology extraction -- the historical pattern through which technological knowledge was taken from Black Americans, repackaged as the inventions of their enslavers, and written into a national narrative from which its originators were systematically erased. I argued that this pattern did not end with emancipation or industrialization. It is alive in the AI era, and superintendents who do not understand it will repeat it.
Let me be specific about how.
The labor that powers artificial intelligence is disproportionately performed by people of color in the Global South. The training data that makes large language models functional is annotated, labeled, and moderated by workers in Kenya, the Philippines, India, and other nations where labor costs are low enough to make the economics of AI profitable (Atanasoski & Vora, 2019). These workers -- who read and categorize the content that teaches AI systems to generate coherent text, identify objects in images, and filter harmful material -- are paid a fraction of what their labor produces. Their intellectual contribution is erased in the marketing of AI products, which are presented as feats of Silicon Valley engineering rather than products of a global labor system that mirrors the extraction patterns Trotter documented two centuries ago.
This is relevant to education because the AI tools entering classrooms carry this history in their architecture. When a superintendent approves the purchase of an AI tutoring platform, that platform's capabilities rest on a foundation of extracted labor. When a teacher uses an AI tool to generate culturally responsive lesson plans, the tool's ability to do so depends on whether the data workers who trained it were themselves from culturally diverse backgrounds and whether the training process valued cultural knowledge -- or treated it as noise to be normalized away. When a student interacts with an AI chatbot, the chatbot's language patterns, its assumptions about what constitutes "correct" expression, its range of cultural references, all reflect decisions made by teams and training processes that the student, the teacher, and the superintendent have no visibility into.
Sinclair's (1998) framework asks us to examine technology through four phases: how it is imagined, produced, employed, and experienced. Applied to AI in education, the equity questions at each phase are urgent:
Imagined: Who decided that this AI tool was needed? Whose problem does it solve? Were communities of color involved in defining the problem, or only in being the population the solution was designed to serve? When edtech companies imagine their products, they imagine users -- and those imagined users, overwhelmingly, look like the engineers and venture capitalists who fund the work, not the Black and Latino students in high-poverty districts who will be most affected by the tool's assumptions.
Produced: Who built the tool? Whose data trained the model? Were the training datasets representative of the linguistic, cultural, and experiential diversity of the students who will use the product? Were bias audits conducted -- not as a compliance checkbox but as a genuine interrogation of how the tool performs across racial, linguistic, and socioeconomic lines?
Employed: Who has access to the tool, and under what conditions? Are the districts deploying AI the same districts that have trained their teachers to use it critically, or are high-poverty districts receiving technology without capacity -- the educational equivalent of shipping textbooks to schools without training teachers to teach from them?
Experienced: How do students from different racial, linguistic, and socioeconomic backgrounds actually experience the tool? Are outcomes disaggregated? Are students asked? Are families informed? Is there a mechanism for communities of color to shape, challenge, or reject the technology if it is not serving their children?
These are not abstract questions. They are the questions that should be on every superintendent's desk before any AI contract is signed. And they are the questions that, in my experience, are almost never asked.
Extending Khalifa's Culturally Responsive School Leadership to AI
Muhammad Khalifa's (2018) Culturally Responsive School Leadership was a landmark contribution to the field, and his framework has shaped how an entire generation of school leaders thinks about equity. Khalifa identified four essential strands of culturally responsive school leadership: (1) critical self-awareness of one's own identity, biases, and positionality; (2) developing culturally responsive teachers through hiring, professional development, and accountability; (3) promoting inclusive school environments that affirm students' cultural identities; and (4) engaging students' community and home contexts as assets rather than deficits.
No one, to my knowledge, has systematically extended Khalifa's CRSL framework to the specific context of AI adoption. This is the bridge this chapter builds -- the connection between culturally responsive leadership and the Fourth Industrial Revolution that no other book has made.
[FIGURE 9.2: Khalifa's CRSL Framework Extended to AI -- Four-column table. Column 1: CRSL Strand. Column 2: Original Application (Khalifa, 2018). Column 3: AI-Era Extension. Column 4: Superintendent Action.
Row 1: Critical Self-Awareness | Leaders examine their own racial identity and biases | Leaders examine their assumptions about AI neutrality and their comfort delegating decisions to algorithms | Conduct a personal AI bias audit: What do you assume AI tools do well? For whom? Based on what evidence?
Row 2: Culturally Responsive Teachers | Hire and develop teachers who affirm students' cultural identities | Train teachers to evaluate AI tools for cultural bias, not just functionality | Require AI professional development that includes bias literacy, not just technical how-to
Row 3: Inclusive Environments | Create schools where students see themselves reflected | Ensure AI tools reflect students' languages, cultures, and epistemologies | Require vendors to provide demographic data on training datasets and bias audit results
Row 4: Community Engagement | Center families and community knowledge | Include communities of color in AI adoption decisions, not just as end-users | Establish community AI advisory committees with decision-making authority, not advisory role only]
Critical Self-Awareness in the AI Context. Khalifa's first strand requires leaders to examine their own identities, biases, and positions of power. In the AI context, this means examining the assumptions we carry about technology's neutrality. I have sat in rooms with superintendents who believe, genuinely and without malice, that AI tools are objective because they are mathematical. They are not. Every algorithm embeds the values and blind spots of the people who designed it and the data that trained it. A superintendent who has not interrogated their own assumption of AI neutrality cannot lead equitable AI adoption, because they cannot see what needs to be questioned.
The superintendent I studied for my dissertation modeled this kind of critical self-awareness when he insisted on naming the specific student populations his theory of action was designed to serve. "You must name it," he said. "When it's not named, and everything is meshed together, then you don't have a specific game plan" (Martin, 2022, p. 82). Khalifa (2018) would recognize this as culturally responsive leadership in action -- the refusal to hide behind universal language that serves no one specifically. In the AI context, the equivalent act of naming is this: Which students does this AI tool serve well, and which students does it underserve? Show me the disaggregated data.
Culturally Responsive Teachers in the AI Context. Khalifa's second strand focuses on developing teachers who can affirm students' cultural identities in their pedagogy. In the AI era, this means teachers must be equipped not only to use AI tools but to interrogate them -- to recognize when an AI's recommendations reflect cultural bias, when its language corrections penalize non-dominant dialects, when its personalization pathways replicate tracking patterns. This is a fundamentally different kind of professional development than what most districts are providing. The RAND data shows that districts are training teachers on how to use AI tools. They are not, by and large, training teachers on how to evaluate those tools for equity -- how to ask the vendor about training data demographics, how to spot differential recommendation patterns, how to center student voice in assessing whether a tool is culturally affirming or culturally erasive.
Inclusive Environments in the AI Context. Khalifa's third strand calls for school environments where students see themselves reflected -- in the curriculum, in the adults, in the physical space, in the culture. AI tools are now part of that environment. When an AI writing assistant does not recognize African American Vernacular English as a legitimate linguistic system, it sends a message about whose language matters. When an AI tutoring platform's reading recommendations center White, middle-class narratives, it tells students of color that their experiences are peripheral. When AI-generated images used in educational materials consistently default to White faces unless specifically prompted otherwise, it reproduces the erasure that Sinclair and Trotter documented in the history of technology itself. An inclusive environment in 2026 requires inclusive technology -- and that requires asking vendors questions they are not accustomed to answering.
Community Engagement in the AI Context. Khalifa's fourth strand centers families and communities as essential partners in education, not as passive recipients of decisions made by professionals. The superintendent I studied modeled this through parent affinity groups -- the African American Parent Advisory Committee, the Latino Parent Alliance -- that were given genuine voice in shaping district direction. He described his approach as "sell, don't tell": take community members to see the work in action rather than lecturing them about why it matters. In the AI context, community engagement means including families and community organizations in AI adoption decisions. It means asking: Do the families in our community understand what AI tools are doing with their children's data? Do they have a voice in whether those tools are used? Do they have the power to say no?
This is not a theoretical exercise. The Brookings Institution's 2025 global study on AI in education, spanning more than 50 countries, concluded that "the risks of utilizing generative AI in children's education overshadow its benefits" given current patterns of use (Brookings Institution, 2025). When risks overshadow benefits, the communities bearing those risks must be at the decision-making table -- not as an afterthought, but as a condition of adoption.
Naming What Is Operating: A CRT Analysis of AI in Education
Throughout this book, I have used Critical Race Theory not as a political position but as an analytical lens -- a way of seeing what is operating beneath the surface of systems that present themselves as neutral. CRT insists on several principles that are directly relevant to AI adoption in schools.
First, CRT holds that racism is ordinary, not aberrational -- that it is the everyday experience of most people of color in America, embedded in the normal operations of institutions rather than limited to individual acts of prejudice (Delgado & Stefancic, 2017). Applied to AI: algorithmic bias is not a malfunction. It is the ordinary operation of systems trained on data generated by a society structured by racial hierarchy. An AI that reproduces inequity is not broken. It is working exactly as its training data dictates.
Second, CRT insists on interest convergence -- the principle that advances in racial equity tend to occur only when they align with the interests of the dominant group (Bell, 1980). Applied to AI in education: districts are most likely to invest in equitable AI when doing so also serves the interests of their most privileged constituents. The superintendent who frames equitable AI as beneficial for all students -- as the Innovation USD superintendent did when he said, "You helping Blacks is for everyone. You helping Latinos is for everyone" (Martin, 2022, p. 85) -- is practicing interest convergence strategically, building support for equity-centered policy by demonstrating its universal benefit.
Third, CRT demands counter-storytelling -- centering the narratives of people of color against dominant narratives that normalize their exclusion (Solorzano & Yosso, 2002). Applied to AI: whose stories does the AI tell? Whose knowledge does it treat as authoritative? When a student asks an AI chatbot about American history and receives a narrative centered on European achievement, that is not a neutral response. It is a counter-story suppressed. When an AI career counseling tool steers a Black student toward vocational paths based on "predictive" data shaped by decades of racialized tracking, that is not personalization. It is the automation of a narrative that says certain children are meant for certain futures.
The questions I am raising here are not comfortable. They are not meant to be. They are meant to do what CRT has always done: make visible the structures that operate most effectively when they are invisible. AI's greatest danger in education is not that it will fail. It is that it will succeed -- at reproducing the exact inequities it was deployed to address, with the added authority of algorithmic objectivity making those inequities harder to name, harder to challenge, and harder to dismantle.
What Equitable AI Adoption Actually Looks Like
I have spent this chapter naming what is wrong. Let me now name what is possible.
Equitable AI adoption is not a prohibition on technology. It is a discipline of inquiry -- a set of practices that ensure every AI tool entering a district has been evaluated not only for functionality and cost but for its impact on the students who have historically been served last and least. It is, in the language of my framework, the application of Sinclair's four phases with equity checkpoints at every transition.
Based on my research, the literature on culturally responsive leadership, and the emerging evidence on AI in education, I propose the following Equitable AI Adoption Protocol -- a structured decision-making process organized by Sinclair's four phases, designed to be used by superintendents and their cabinets before any AI tool or platform is adopted, piloted, or scaled.
[FIGURE 9.3: The Equitable AI Adoption Checklist -- One-page visual checklist organized by Sinclair's four phases.
IMAGINE Phase: Before You Decide
- Who identified the need this tool addresses? Were communities of color involved in defining the problem?
- Does the tool's design reflect the cultural, linguistic, and experiential diversity of our students?
- Have we examined our own assumptions about AI neutrality? (Khalifa Strand 1)
- What problem are we solving, and for whom? Name the specific student populations.
PRODUCE Phase: Before You Purchase
- What data trained this model? Request the vendor's training data demographics.
- Has the tool been independently audited for racial, linguistic, and socioeconomic bias?
- Does the vendor employ a diverse design team? Ask for their DEI data.
- Will the tool work equitably for students with limited broadband, older devices, or non-dominant language backgrounds?
EMPLOY Phase: Before You Deploy
- Have all teachers received training that includes AI bias literacy, not just technical functionality? (Khalifa Strand 2)
- Is access equitable across schools, or are well-resourced schools getting the tool first?
- Have families been informed and given genuine voice in the adoption decision? (Khalifa Strand 4)
- Is there a plan to monitor disaggregated outcomes from day one?
EXPERIENCE Phase: After Deployment
- Are outcomes disaggregated by race, language, socioeconomic status, and disability?
- Are students being asked directly about their experience with the tool? (Center student voice)
- Is the tool creating differential pathways that replicate tracking? Check recommendation patterns.
- Is there a mechanism for communities to challenge, modify, or reject the tool based on evidence?
- Review cycle: Quarterly equity check-ins using disaggregated data, with community participation.]
This protocol is not a bureaucratic obstacle to AI adoption. It is a discipline of care -- the same discipline that the superintendent of Innovation USD practiced when he insisted on naming the students his policies were designed to serve, when he created parent affinity groups with real voice, when he conducted quarterly equity check-ins on every school's progress. The protocol simply applies that same discipline to technology decisions.
Let me be direct: most AI vendors will not be able to answer these questions. They will not have demographic data on their training datasets. They will not have conducted independent bias audits. They will not have disaggregated outcome data by race. This is not a reason to abandon the protocol. It is a reason to use it. When a vendor cannot answer basic equity questions about their product, that tells you something essential about the product -- and about the assumptions embedded in its design.
The OECD (2025) has called for education systems to play a central role in "ensuring equitable AI" and "shaping an inclusive digital future." Digital Promise (2024) has identified leadership for digital transformation as the first of five key domains for equitable AI in education. The British Educational Research Association (2024) has framed digital equity in the age of generative AI as requiring systemic rather than individualized responses. The research consensus is clear: equitable AI is a leadership responsibility, not a technology department function. It belongs on the superintendent's desk.
The Moral Case: Who Designs, Who Benefits, Who Is Surveilled
I want to close this chapter with three questions that I believe should guide every superintendent's thinking about AI and equity. They are drawn from the tradition of Critical Race Theory, from Sinclair's framework, from Khalifa's CRSL, and from my own experience as a Black woman who has spent a career watching well-intentioned technology initiatives reproduce the hierarchies they claimed to dismantle.
Who designs the algorithms? As of 2023, women constituted less than 30% of global AI talent (OECD, 2025). Racial minorities are even more dramatically underrepresented. The people designing the tools that will shape how 50 million American students learn do not, by and large, look like those students, speak their languages, or share their cultural referents. This is not an incidental feature of the AI industry. It is a structural condition that produces tools encoded with specific assumptions about intelligence, language, behavior, and potential -- assumptions that favor the communities from which the designers come and disadvantage the communities they do not know.
Whose data trains the models? Every AI system is a product of its training data. When that data is drawn overwhelmingly from the Global North, from English-language sources, from contexts that center Whiteness as the norm, the system's outputs will reflect those origins -- in what it recommends, in how it evaluates, in what it treats as "correct" and what it flags as deviant. Frontiers in Computer Science (2026) documented that AI systems trained on Global North datasets "often fail to reflect the linguistic, cultural, and contextual needs of diverse populations." This is not a technical limitation. It is a design choice with equity consequences.
Who benefits and who is surveilled? In affluent districts, AI is positioned as a tool of empowerment -- personalized learning, creative extension, student agency. In under-resourced districts, AI is more likely to be positioned as a tool of control -- monitoring behavior, flagging cheating, predicting which students will drop out. The Center for Democracy and Technology's 2025 report found that schools' embrace of AI was connected to increased risks to students, including deepfake threats, AI companion dependencies, and surveillance through data collection. These risks are not distributed equally. Students in under-resourced districts, who are disproportionately Black and Brown, are more likely to experience AI as a surveillance mechanism and less likely to experience it as a creative tool. This is Trotter's technology extraction in digital form: the benefits flow to those who already have power, while the costs and risks are borne by those who do not.
The Fourth Industrial Superintendent sees these patterns because they have done the work Khalifa demands: examined their own assumptions, built culturally responsive teams, created inclusive environments, and centered community voice. They see AI not as a neutral tool but as a technology with a history -- a history of extraction, a history of exclusion, and, if we choose it, a possible history of transformation.
But that transformation will not happen by default. It will happen only if superintendents have the moral courage to ask the hard questions, the equity literacy to interpret the answers, and the practical tools to act on what they learn. The protocol in this chapter is one such tool. Your moral purpose is another. The students in your district -- the ones you can name, the ones whose communities have been promised transformation before and received extraction instead -- are the reason both exist.
Discussion Questions
Practitioner Tool: The Equitable AI Adoption Protocol
Purpose: A structured decision-making process for superintendents and district leadership teams evaluating AI tools or platforms, organized by Sinclair's (1998) four phases of technology impact on marginalized communities. Each phase includes equity checkpoints, specific questions to ask vendors, data to collect, and community engagement requirements.
How to Use: Complete the protocol collaboratively with your cabinet, technology director, equity officer, and community advisory committee before any AI tool is piloted or adopted district-wide. Score each phase on a 1-4 scale (1 = not addressed, 2 = partially addressed, 3 = substantially addressed, 4 = fully addressed with community input). A total score below 10 (out of 16) indicates the tool is not ready for equitable deployment.
Phase 1: IMAGINE (Score: __ /4)
| Checkpoint | Evidence Required |
|---|---|
| The need for this tool was identified with input from communities of color, not only district staff or vendor marketing | Meeting notes, community forum records, survey data |
| The problem this tool solves has been named for specific student populations, not described in universal terms | Written problem statement naming populations |
| District leadership has examined its own assumptions about AI neutrality (Khalifa Strand 1) | Professional development records, reflection protocols |
| The driving question for adoption centers equity: "Who benefits and who may be harmed?" | Documented equity impact analysis |
Phase 2: PRODUCE (Score: __ /4)
| Checkpoint | Evidence Required |
|---|---|
| The vendor has provided demographic data on training datasets | Vendor documentation |
| Independent bias audit results are available (not only vendor self-assessment) | Third-party audit report |
| The design team includes racial, linguistic, and gender diversity | Vendor DEI data |
| The tool has been tested with students from diverse linguistic and cultural backgrounds | Pilot data disaggregated by demographics |
Phase 3: EMPLOY (Score: __ /4)
| Checkpoint | Evidence Required |
|---|---|
| Teacher training includes AI bias literacy, not only technical functionality (Khalifa Strand 2) | PD curriculum and agendas |
| Deployment plan ensures equitable access across all schools, not privileging well-resourced sites first | Rollout timeline with equity sequencing |
| Families have been informed and given decision-making voice, not only notification (Khalifa Strand 4) | Community engagement records, translation evidence |
| Disaggregated monitoring plan is in place from day one | Data collection protocol |
Phase 4: EXPERIENCE (Score: __ /4)
| Checkpoint | Evidence Required |
|---|---|
| Outcome data is disaggregated by race, language, SES, and disability at every review | Quarterly data reports |
| Students are asked directly about their experience with the tool | Student voice survey or focus group data |
| AI recommendation and pathway patterns are monitored for differential tracking | Algorithmic audit reports |
| Community mechanism exists to challenge, modify, or discontinue the tool | Published process with documented decisions |
Total Score: __ /16
- 13-16: Proceed with confidence; maintain quarterly equity check-ins
- 10-12: Proceed with caution; address gaps before scaling beyond pilot
- Below 10: Do not deploy; return to Phase 1 and rebuild with community input
References
Atanasoski, N., & Vora, K. (2019). Surrogate humanity: Race, robots, and the politics of technological futures. Duke University Press.
Bell, D. A. (1980). Brown v. Board of Education and the interest-convergence dilemma. Harvard Law Review, 93(3), 518-533.
Benjamin, R. (2019). Race after technology: Abolitionist tools for the New Jim Code. Polity Press.
British Educational Research Association. (2024). Digital equity in the age of generative AI: Bridging the divide in educational technology. https://www.bera.ac.uk/blog/digital-equity-in-the-age-of-generative-ai-bridging-the-divide-in-educational-technology
Brookings Institution. (2025). A new direction for students in an AI world: Prosper, prepare, protect. https://www.brookings.edu/articles/a-new-direction-for-students-in-an-ai-world-prosper-prepare-protect/
Buolamwini, J., & Gebru, T. (2018). Gender shades: Intersectional accuracy disparities in commercial gender classification. Proceedings of Machine Learning Research, 81, 1-15.
Center for Democracy & Technology. (2025). Hand in hand: Schools' embrace of AI connected to increased risks to students. https://cdt.org/insights/hand-in-hand-schools-embrace-of-ai-connected-to-increased-risks-to-students/
Crenshaw, K. (1989). Demarginalizing the intersection of race and sex: A Black feminist critique of antidiscrimination doctrine, feminist theory and antiracist politics. University of Chicago Legal Forum, 1989(1), 139-167.
de la Pena, C. (2010). The history of technology, the resistance of archives, and the Whiteness of race. Technology and Culture, 51(4), 919-937.
Delgado, R., & Stefancic, J. (2017). Critical race theory: An introduction (3rd ed.). New York University Press.
Digital Promise. (2024). How AI for education can address digital equity. https://digitalpromise.org/2024/02/20/how-ai-for-education-can-address-digital-equity/
Federal Communications Commission. (2024). Broadband deployment report. https://www.fcc.gov/reports-research/reports/broadband-progress-reports
Frontiers in Computer Science. (2026). AI and the digital divide in education. https://www.frontiersin.org/journals/computer-science/articles/10.3389/fcomp.2026.1759027/full
Frontiers in Education. (2025). Empowering educational leaders for AI integration in rural STEM education: Challenges and strategies. https://www.frontiersin.org/journals/education/articles/10.3389/feduc.2025.1567698/full
Khalifa, M. A. (2018). Culturally responsive school leadership. Harvard Education Press.
Martin, M. G. (2022). The Fourth Industrial Superintendent [Doctoral dissertation, University of Southern California]. USC Digital Library.
National Center for Education Statistics. (2024). Rural students' access to the internet. Condition of Education. https://nces.ed.gov/programs/coe/indicator/lfc/internet-access-students-rural
Noble, S. U. (2018). Algorithms of oppression: How search engines reinforce racism. New York University Press.
OECD. (2024). The potential impact of artificial intelligence on equity and inclusion in education (OECD Digital Economy Papers, No. 23). OECD Publishing. https://www.oecd.org/en/publications/the-potential-impact-of-artificial-intelligence-on-equity-and-inclusion-in-education_15df715b-en.html
OECD. (2025). Ensuring equitable AI: The role of education in shaping an inclusive digital future. OECD Blog. https://www.oecd.org/en/blogs/2025/04/ensuring-equitable-ai-the-role-of-education.html
RAND Corporation. (2025). More districts are training teachers on artificial intelligence (RR-A956-31). https://www.rand.org/pubs/research_reports/RRA956-31.html
Selwyn, N. (2022). The future of AI and education: Some cautionary notes. European Journal of Education, 57(4), 620-631.
SETDA. (2025). Sustaining progress to close the digital access divide in K-12 education. https://www.setda.org/wp-content/uploads/2025/01/SETDA_UCI-Report-2025_Official.pdf
Sinclair, B. (1998). Teaching about technology and African American history. OAH Magazine of History, 12(2), 14-17.
Solorzano, D. G., & Yosso, T. J. (2002). Critical race methodology: Counter-storytelling as an analytical framework for education research. Qualitative Inquiry, 8(1), 23-44.
The Markup. (2025). AI detection tools and non-native English speakers. https://themarkup.org/
Trotter, J. (2000). African Americans and the Industrial Revolution. OAH Magazine of History, 15(1), 19-23.
World Economic Forum. (2025). The Future of Jobs Report 2025. https://www.weforum.org/publications/the-future-of-jobs-report-2025/