Building Digital Infrastructure for Black Learners in the Age of AI

“We are talking about a design problem rooted in exclusion. Bias is engineered into the logic of the systems and tools we use in EdTech and education.”

Hassanatu Blake, PhD, Co-Founder & President at ED2Tech, Associate Clinical Professor and Inaugural Director of the Global Health Undergraduate Program at the UM School of Public Health

As artificial intelligence becomes embedded across edtech platforms, concerns about bias are often addressed at the level of outputs—whether systems produce fair results, whether datasets are sufficiently representative, or whether technical adjustments can reduce disparities. But that focus misses where many of these issues originate.

In education, AI systems are not neutral tools layered onto existing environments; they are built within institutional contexts that shape how students are evaluated, categorized, and supported from the outset. The result is that inequities can be introduced long before a model generates a result, embedded in the assumptions, definitions, and decision-making frameworks that guide how these systems function.

Work from researchers and practitioners in this space points to a more structural dynamic. Bias is not only a question of flawed data or imperfect algorithms, but of how educational technologies define concepts such as risk, performance, and engagement, and whose experiences inform those definitions. When those reference points are narrow, systems can misread or penalize students whose lived realities fall outside those assumptions, reinforcing patterns that predate the technology itself.

This shifts the conversation away from access and toward control. Expanding access to devices or increasing representation within existing institutions may improve participation, but those approaches do not necessarily change who determines how systems are built or what outcomes they prioritize. As AI becomes more central to how education is delivered and governed, the more consequential question is whether these systems will continue to reflect inherited structures, or whether new forms of digital infrastructure can be developed that embed community authority, accountability, and long-term ownership into their design.

To better understand how these structural dynamics are unfolding in practice, we spoke with Dr. Symone Campbell, Raymar Hampshire, Dr. Hassanatu Blake, and Hussainatu Blake, who recently spoke on the panel Building Digital Infrastructure for Black Learners and whose work focuses on how design processes, data governance, and institutional power shape the development and impact of AI in education.

Dr. Hassanatu Blake
Hassanatu Blake, PhD, MPH, MBA is co-founder/president at ED2Tech and associate clinical professor and inaugural director of the global health undergraduate program at the University of Maryland School of Public Health. She advises on digital innovation, global learning, and policy.

Dr. Symone Campbell
Symone Campbell, PhD is an internal resident fellow at Siegel Family Endowment and adjunct professor of AI ethics at Howard University in the Data Science and Human Development Departments.

Raymar Hampshire
Raymar Hampshire is the founder and product developer of Braid, an open-source vocational storytelling platform and knowledge network. He is also research lead at the Public Interest Technology Knowledge Network (PIT-KN) lab at the University of Michigan. His work explores human-centered storytelling and sociotechnical systems.

Hussainatu Blake
Hussainatu Blake, JD, MA is CEO of ED2Tech, an education and workforce consulting firm helping organizations responsibly harness AI and emerging technologies. She works at the intersection of education and technology to build the next generation of learning systems.

In current AI and edtech systems, bias does not emerge as an isolated flaw. It reflects how these systems are built, what data they rely on, and the assumptions embedded in how students are evaluated. Focus must move away from outputs alone and toward the underlying structures that shape how those outputs are produced.

Dr. Hassanatu Blake explains, “We are talking about a design problem rooted in exclusion. Bias is engineered into the logic of the systems and tools we use in edtech and education.” The consequences are already visible in how these tools operate.

“Predictive analytics systems have flagged Black students as ‘high risk’ at four times the rate of white peers,” Blake notes, while “AI proctoring tools struggle with darker faces, which leads to mistaken identities and harsh academic penalization.” These outcomes are not separate from system design; they follow directly from how tools are trained, tested, and deployed. When technologies are developed without reflecting the communities most affected by them, they can “encode and amplify the racial biases already historically embedded in our societies.”

Raymar Hampshire situates this more broadly, describing that “AI acts as an accelerator for existing biases.” In education, that acceleration is shaped by both data and assumptions about learners. He points to a familiar example: standardized test questions that assume prior knowledge or exposure, such as knowing what a yacht is. AI and edtech systems operate within similar constraints, where learning outcomes are shaped by embedded expectations about students’ lived experiences.

When systems are designed without incorporating a wider range of experiences and perspectives, they do not simply produce uneven results. They reproduce existing disparities in ways that can scale across classrooms, districts, and platforms. Addressing bias, in this context, requires more than technical adjustment. It requires rethinking how these systems are designed from the outset, including who is involved in shaping them and what assumptions guide their development.

Discussions of equity in edtech frequently emphasize access and participation: device availability, broadband connectivity, and representation in hiring or advisory roles. These measures expand who can engage with existing systems, but they operate within structures where core decisions about design, standards, and data remain centralized. The speakers point to a different layer of the problem—one that sits upstream of access itself.

As Dr. Hassanatu Blake explains, many of these efforts are “necessary but insufficient because they focus on existing power structures without challenging them.”

The limitation is structural. When authority over how technologies are built and governed remains unchanged, participation does not extend to determining how those systems function or whose needs they prioritize. The question becomes less about who is included in a system and more about who defines its terms.

Black digital infrastructure is articulated as a response at that level. Blake describes it as “creating an ecosystem for communities to participate as creators, decision-makers, and beneficiaries of edtech,” with an emphasis on “community ownership, governance, and data sovereignty.”

Conditions shape long-term outcomes: how knowledge is produced, how data is governed, and how standards are set and enforced. Authority is distributed across these dimensions rather than concentrated within institutions that operate at a distance from the communities most affected.

That approach is reflected in how some models are being built. Dr. Blake points to ED2Tech’s co-design work, which “brings communities in as knowledge producers” and supports Black founders in moving “through the technology valley to full integration.”

The structure of that process matters. Communities are involved at the point where problems are defined and solutions are developed, shaping both the direction of the technology and the criteria by which it is evaluated. At the same time, building infrastructure requires continuity—pathways that allow ideas, research, and products to move into sustained use rather than remaining isolated interventions.

At the same time, governance in AI-driven education is developing alongside the systems it aims to regulate. In many cases, that development takes the form of principles and guidelines rather than enforceable standards. Institutions adopt tools, companies define internal practices, and policy frameworks begin to emerge, but the mechanisms that translate expectations into consistent outcomes remain uneven.

Across these dimensions, governance takes shape through a combination of standards, participation, and material support. As AI becomes more integrated into educational systems, these factors influence how consistently those systems operate and whose priorities are reflected in their expansion.

“Voluntary principles are insufficient because they only signal good intent,” says Hussainatu Blake, highlighting one of the main issues.

Without clear standards, defined timelines, and independent oversight, enforcement depends on how individual actors interpret their responsibilities. In practice, that leaves room for inconsistency across platforms, districts, and use cases, even as these systems take on a more central role in shaping educational decisions.

Also, who participates in setting those standards continues to shape how governance operates. “Black edtech leaders represent only 2 percent of the sector, yet we serve the communities most impacted by these tools,” Blake notes. The composition of decision-making bodies influences how risks are defined, which issues are prioritized, and how solutions are structured. When representation is limited at that level, governance can reflect a narrower set of assumptions about how systems function in practice.

Evaluation introduces a different constraint. Systems can meet performance benchmarks while still producing outcomes that do not align with the communities they affect.

Dr. Symone Campbell highlights the need for “ongoing impact audits that assess not only technical performance but social and cultural outcomes.” Measuring those dimensions requires a broader definition of system performance, extending beyond accuracy or efficiency to encompass how decisions are experienced and interpreted.

Control over data raises additional questions about how value is created and distributed. On this point, Hussainatu Blake points out that large technology companies have “been scraping culturally responsive data without consent, attribution, or compensation.”

The way data is sourced and used shapes who benefits from its application. Where protections are limited, incorporating that data into AI systems can separate knowledge from the communities that produce it. Meanwhile, resource allocation operates alongside these dynamics.

“Governance without resource allocation is just rhetoric,” Blake notes. Capital influences which approaches are sustained over time, which organizations are able to scale, and how standards translate into practice. The presence of a framework does not determine its reach; the availability of resources often does. How those resources are structured and sustained becomes especially important as these models move beyond initial development and into broader adoption.

Scaling community-centered approaches in EdTech introduces a different set of constraints. Growth often requires alignment with funding structures, institutional partnerships, and market expectations that can reshape how a model operates as it expands. The challenge is not only reaching a broader audience, but doing so without losing the elements that define how the system was built and governed.

Hussainatu Blake describes scaling as dependent on building “parallel infrastructure,” supported through a combination of funding sources rather than a single pathway.

Philanthropic capital, community development financial institutions, federal small-business and minority innovation funds, and longer-term patient capital each play a role in sustaining models not designed for rapid extraction or short-term returns. The structure of that funding affects how organizations grow and what tradeoffs they face as they expand.

Talent development operates on a similar timeline. Blake emphasizes the need to invest in “Black researchers, technologists, and entrepreneurs” in ways that support “intergenerational knowledge transfer” and allow leaders to “set the agenda” while preparing the next generation of builders. This extends beyond entry-level access into the field. It shapes who has the capacity to design systems, direct research, and influence how technologies evolve over time.

As models gain visibility, the terms under which they engage with larger systems become more consequential.

“Protecting IP, maintaining governance structures, and preserving cultural integrity are non-negotiable,” Blake notes. The ability to scale without dilution depends on whether those conditions are maintained as partnerships form and adoption increases.

Dr. Symone Campbell frames scaling in terms of continuity across an ecosystem rather than expansion of individual organizations. Through initiatives such as The Black EdTech Database, she focuses on strengthening “visibility, connectivity, and shared infrastructure among Black-led and community-centered innovators.” This approach allows models to grow through shared standards, research, and data practices, supporting broader reach while maintaining alignment with their original design principles.

The structure that emerges is distributed rather than centralized. Growth occurs across networks that share resources, collaborate on development, and reinforce common approaches to governance. In this context, scaling reflects how systems are sustained across different environments, not only how widely they are adopted. What carries forward is not a single platform or organization, but a set of conditions-ownership, authority, and continuity-that determine whether these models retain their integrity as they expand.