The Unequal Economics of AI Learning: Why Free Tools Still Come at a Cost

Generative artificial intelligence is rapidly becoming an academic resource, promising students easier access to explanations, writing assistance, research support and personalised learning. Yet beneath its growing adoption lies a fundamental inequality: students who can access the same AI tools do not necessarily have equal opportunities to benefit from them. The difference increasingly depends on connectivity, financial resources, digital competence and the institutional environments surrounding their education.

A new study titled "Generative AI and Digital Inequality in South African Higher Education: Students' Experiences of Access and Meaningful Use at a Rural University," published in the journal Education Sciences, examines this divide. Conducted by Theledi Nkube and Khashane Malatji of the University of Mpumalanga, the research reveals how existing socioeconomic and technological inequalities shape students' ability to translate generative AI access into meaningful academic opportunities.

Drawing on interviews with 12 undergraduate students at the university's Siyabuswa campus, the researchers identify interconnected barriers involving devices, internet reliability, mobile data affordability, AI literacy and institutional support. Their findings challenge the assumption that making AI platforms available automatically democratises education. Instead, technological access can coexist with unequal learning opportunities, creating a divide between students who recognise AI's potential and those equipped to realise it.

Beyond ChatGPT Access: The Hidden Inequalities Shaping AI-Powered Learning

The expansion of generative AI has brought new possibilities into university classrooms. Students increasingly use conversational AI systems to clarify difficult concepts, summarise readings, support academic writing and prepare for examinations. Unlike conventional digital resources, these systems generate responses directly, potentially making educational assistance more accessible to learners who need additional academic support.

However, the South African research challenges the notion that technological availability translates automatically into educational inclusion. Participants described situations in which they possessed smartphones and understood the potential usefulness of AI applications but could not use them consistently because of unstable internet connections or limited mobile data. Their experiences expose differences between having access to a platform and maintaining the conditions necessary to benefit from it.

These distinctions are particularly significant in rural higher education, where students may depend on campus Wi-Fi, shared computers or personal smartphones. The research found that connectivity problems could interrupt assignments, while dependence on institutional networks sometimes determined when students could engage with AI. Students with personal laptops and reliable internet were perceived as having greater opportunities for sustained use.

The findings extend established discussions of digital inequality beyond basic internet availability. Digital participation depends on the quality, continuity and suitability of access, not simply whether a connection exists. Generative AI makes these differences particularly consequential because meaningful engagement can involve repeated questions, clarification, comparison and evaluation of generated responses.

Financial circumstances introduce another dimension. Participants relying on mobile data reported postponing AI use until free Wi-Fi became available or prioritising academic activities when their data allowances were limited. Access therefore operated as an economic condition, influencing the frequency and flexibility of students' engagement with educational technology.

For students facing these constraints, using AI is not necessarily a straightforward decision based on academic need. It can depend on whether sufficient connectivity is available at the required moment. The study consequently suggests that economic disadvantage may restrict opportunities to develop familiarity with AI, even among students who already possess compatible devices.

The AI Literacy Divide: When Equal Technology Produces Unequal Benefits

Material access explains only part of the inequality identified by the researchers. Students also differed in their ability to formulate effective prompts, evaluate generated information and determine whether AI responses were academically reliable. Such differences influenced how confidently and productively participants could engage with the same technological applications.

Some participants described difficulty obtaining useful responses because they initially lacked the ability to ask appropriate questions. Others recognised that students who simply copied generated answers without verification might use AI less critically than those able to interrogate its responses. These experiences demonstrate that technological familiarity does not necessarily produce informed academic judgment.

Generative AI introduces challenges that extend beyond conventional digital literacy. Unlike systems primarily directing users towards existing information, conversational AI can produce confident, plausible explanations containing inaccuracies. Effective academic use therefore requires students to verify claims, recognise limitations and evaluate generated material against credible sources.

The researchers identify AI literacy as a distinct dimension of educational inequality. Students possessing stronger skills in prompt formulation, information evaluation and critical engagement may derive greater academic value from the same application than peers lacking those competencies. Unequal opportunities to practise these skills can reinforce differences already created by connectivity and affordability.

This relationship highlights one of the study's most important contributions: the distinction between perceived usefulness and realised usefulness. Perceived usefulness describes students' belief that AI can support academic work, while realised usefulness concerns their actual capacity to obtain educational benefits under existing financial, material, technical and institutional conditions.

A student may believe that an AI application could improve understanding of a difficult subject but lack sufficient connectivity to use it when required. Another may enjoy reliable access yet struggle to assess the accuracy of generated explanations. In both circumstances, positive attitudes towards technology do not guarantee meaningful academic outcomes.

By extending the Technology Acceptance Model through a digital inequality perspective, the study places structural conditions alongside individual perceptions. Its central argument is that technology adoption cannot be understood independently of the resources and competencies determining whether users can convert access into practical benefits.

Universities Face a Bigger Challenge Than Simply Providing AI Tools

The findings place institutional responsibility at the centre of debates over equitable AI integration. Participants identified reliable campus connectivity, functional computer laboratories, digital resources and structured training as important conditions for more inclusive participation. Their recommendations reflected recognition that individual effort alone cannot overcome every barrier to educational technology use.

Researchers caution that these accounts represent students' perceptions and proposed remedies rather than independent assessments of existing university provision. Nevertheless, they provide insight into the institutional arrangements students consider necessary for meaningful engagement with generative AI.

The study recommends an integrated response combining infrastructure, affordability, AI literacy and academic guidance. Improving campus Wi-Fi without developing students' critical competencies may expand connectivity without ensuring effective use. Similarly, providing training without reliable devices or affordable access may leave students unable to apply newly acquired skills consistently.

AI literacy programmes should therefore extend beyond introductory demonstrations of popular applications. The researchers recommend teaching students how to formulate prompts, assess information quality, identify fabricated or inaccurate outputs, protect personal information and exercise independent judgment when incorporating AI into academic activities.

Teaching and assessment practices also require consideration. Academic staff should avoid assuming that every student possesses equivalent technological resources when incorporating generative AI into coursework. Providing alternatives for students facing connectivity or device constraints could help prevent access inequalities from becoming barriers to completing academic tasks.

Institutional AI policies must also balance academic integrity with educational inclusion. Rules concerned exclusively with misconduct may overlook the unequal circumstances under which students access and use AI systems. The study instead advocates policies that integrate responsible use, privacy protection, teaching support, infrastructure and student access.

Such recommendations carry significance for resource-constrained higher education environments. They suggest that equitable AI adoption requires coordinated institutional planning rather than isolated investments in applications or training. The effectiveness of educational technology depends on the wider system supporting students' ability to use it.

The Next Digital Divide Will Be Measured by Educational Benefit, Not Adoption

The research offers a useful perspective on how universities might evaluate progress towards inclusive AI-supported education. Platform availability, student awareness and reported enthusiasm provide incomplete indicators when learners experience substantial differences in connectivity, affordability and competence.

Its qualitative design gives depth to experiences that broad adoption measures may overlook. Through interviews conducted in March 2026 with students from Education, Agriculture and Economics, the researchers examined how multiple constraints interact in everyday academic activity. Their analysis demonstrates why infrastructure, financial resources and AI literacy cannot be treated as entirely separate concerns.

However, the study's conclusions require careful interpretation. Its 12 participants came from a single rural university campus and had previous experience using generative AI. Students excluded entirely from these technologies were therefore not represented. The findings cannot establish the prevalence of AI inequality across South African higher education.

The research also relies primarily on self-reported experiences rather than independently measured internet performance, data expenditure or academic achievement. Its cross-sectional design cannot determine how students' practices might evolve as institutional policies, technological capabilities and AI-related competencies change.

These limitations identify important directions for further investigation. Larger comparative studies could examine differences between institutions, while research combining student accounts with connectivity measurements and learning assessments could clarify how particular barriers influence educational outcomes. Evaluations of AI literacy programmes and institutional connectivity initiatives could also establish which responses deliver measurable improvements.

Beyond South Africa, the study provides an analytical framework relevant to debates about inclusive digital transformation and quality education. Its findings suggest that institutions serving economically disadvantaged students need to examine not only whether emerging technologies are accessible but also whether learners possess the conditions required to use them effectively.