In the current academic climate, the integration of artificial intelligence into higher education is treated as an inevitability. Yet, while administrative debates often center on academic integrity and automated grading, a more fundamental shift is occurring. A recent playbook from Teach Access and Every Learner Everywhere highlights a critical tension: AI possesses the capacity to either dismantle long-standing barriers for students with disabilities or, if implemented without rigor, codify new forms of exclusion.

The Outlier Trap: When Data Becomes a Barrier

The primary risk of modern AI lies in its reliance on normative datasets. In the mathematical logic of machine learning, characteristics that deviate from the statistical average are often categorized as “outliers”. For the 16% of the global population living with a disability, this is not merely a technical nuance; it is a structural flaw.

AI models are frequently trained on datasets curated by humans, meaning any existing biases or omissions are inevitably reflected in the technology. When systems are trained on “typical” eye contact, speech patterns, or physical movements, they inadvertently penalize those with neurodivergence or motor impairments. This “ableist bias” is frequently baked into the design of proctoring software and hiring algorithms, turning what should be an objective tool into a gatekeeper that favors a narrow definition of the “standard” student.

A Recommendation for Higher Education: The AI Accessibility Procurement Scorecard

To move from theory to practice, institutions should adopt a standardized method for evaluating third-party AI vendors. This scorecard is recommended for procurement officers and IT departments to quantify a tool’s commitment to inclusion before it enters the campus ecosystem.

Scoring Guide (0–5):

• 0: No evidence/Non-functional.
• 1–2: Partial or manual workarounds required.
• 3: Full compliance with standard expectations.
• 4–5: Proactive features that exceed standards or lead the industry.

Target Score: 20/30 for standard adoption.

Category	Evaluation Criteria	Reputable Reference
Technical Compliance	POUR Principles: Does the interface meet standards for being Perceivable, Operable, Understandable, and Robust?	W3C Web Accessibility Guidelines
Algorithmic Fairness	Bias Mitigation: Has the model been audited for discrimination against non-normative speech or “atypical” inputs?	NIST AI Risk Management Framework
Transparency	Disclosure of Limitations: Does the vendor provide a VPAT (Voluntary Product Accessibility Template) or disclose known model limitations?	Section 508 VPAT Guidelines
Interoperability	AT Compatibility: Is the tool verified to function with screen readers, eye-tracking, or switch controls?	UN Convention (CRPD)
Adaptability	Cognitive Support: Can the AI simplify dense text into “plain language” or offer executive function support?	WHO World Report on Disability
Inclusion	Direct Involvement: Were people with disabilities (PWD) involved in the product’s research, design, and testing?	UNESCO Recommendation on the Ethics of AI

---

Pathways for the Global South: Pragmatism and Sovereignty

For institutions in the Global South, the challenge is compounded by limited infrastructure and the dominance of Western-centric data. However, the shift toward AI offers distinct pathways to enforce accessibility even with limited resources.

1. Leveraging “Small AI” and Mobile-First Solutions

High-bandwidth, cloud-dependent AI is often a barrier in itself. The Global South can prioritize “Small AI”—lightweight models that run locally on mobile devices or offline.

• Pathway: Focus on mobile-first applications that provide text-to-speech or navigation support without requiring constant internet connectivity.
• Further Reading: World Bank: Small AI, Big Impact

2. Agricultural and Maternal Health Accessibility

Accessibility in the Global South often translates to survival and economic stability.

• Agricultural Accessibility: AI-powered assistants in local dialects can bridge the literacy gap for farmers, providing real-time pest management or market pricing.
• Maternal Health: AI-guided handheld ultrasound devices act as a “second set of eyes” for local midwives, identifying complications early in remote areas where doctors are scarce.

3. Strategic South-South Cooperation

Rather than relying on Global North providers, developing nations can share datasets that reflect local languages and cultural nuances.

• Pathway: Collaborative regional hubs can pool resources to build vernacular models, ensuring accessibility tools are not lost in translation.
• Further Reading: UN Office for South-South Cooperation (UNOSSC)

The “Pragmatic Hybridity” Conclusion

Recently, ChatGPT introduced a lite version called ChatGPT Go for developing countries that will cost significantly less. Even with introduction to such “lite” flavors of AI platforms, a critical question remains: Does using global tools like ChatGPT contradict the push for local, sovereign AI? 

The answer lies in Pragmatic Hybridity. For many students in low-resource environments, specialized AI tiers designed for lower data consumption serve as an immediate “bridge” to inclusion. They provide the transcription, summarization, and translation these learners need today.

However, this is a transitional step. The long-term goal is Digital Sovereignty: moving from being consumers of Western AI to creators of localized systems. By using existing tools to solve immediate accessibility gaps while simultaneously building local data capacity, the Global South ensures that students with disabilities are not left behind during the transition to a more representative digital future.