AI in Education: Impairing the Universal Right to Education

Shahyan Naeem is a final year law student at the University of London International Programmes and is currently working as a researcher in the human rights cell at Walker Martineau Saleem Advocates & Legal Consultants.

Introduction 

The advent of modern technology has transformed methods of conducting one’s life. In the sphere of education, Large Language Models (LLMs) powered by Artificial Intelligence (AI)have a pivotal role to play in contemporary times. It would be wrong for one to deny the benefits of such technologies in the realm of education. However, it is worthy to note that AI chatbots, such as GPT-4 and Perplexity AI, while being heavily utilized for educational purposes, sit ill with the fundamental right to education as enshrined in international human rights instruments. In particular, these language-based AI models provide Euro-centric or Western oriented views that remain prevalent in pedagogy, eliminating the presence of different cultures and perspectives in this field. Dissemination of dominant views undermine the universal values inherent to the right to education. Notwithstanding the above, the continued usage of such means to attain education directly infringes elements of accessibilityand acceptability that lay at the core of the right to education’s universality. The impediments to universalism in the field of education warrant changes of attitudes and a human rights approach toward emerging technologies.  

AI-chat bots

AI-models provide user specific responses by the utilization of data received through multiple sources. Such data includes hypotheses and statistical analysis however, the primary source of information comes from the internal database of the chat bot. Operation of such models is thus, reliant on the data stored in them, and this data collection takes place almost always in the West and/or European states. In the contemporary world, strides in the field of AI and technology are, to a great extent, attributable to the US and China. Following the US and China, major developments in AI & Tech are taking place in the EU, India, and Singapore. Since AI primarily functions on the data fed to it, and because these advancements are exclusively concentrated within a few countries, AI systems are susceptible to reflecting the biases inherent in these societies. Furthermore, such databases amplify common narratives in fields, excluding evolving matters and preventing alternative perspectives from being presented. Thus, as the data fed into databases is Euro-centric or dominant in nature, it will be prone to contravene the notion of universalism. 

Universalism and Human Rights

Human rights were built on the foundation of all humans being “free and equal in dignity and rights” thereby entitling equal treatment. Human dignity, at its core, holds that every human has a secure connection with the concept of humanity without distinction of class, race, gender, etc. Such a foundation evidences a ‘universalist’ standpoint – implying that norms and human rights ensured under international jurisprudence will be applied to all regardless of where they may be located.

Many have argued to the contrary as well, believing that rights are to be applied in a manner that conforms to the specific cultures or societies of humans. Conflicts between norms of international law and culture have paved room for ‘cultural relativism’ preventing equal application of rights. However, the concept of relativism has almost always been advanced to trump rights and entitlements of others, and more specifically in South Asian societies, of women. Therefore, the relativist standpoint is inherently flawed.

With respect to the right to education, its presence in international jurisprudence is ensured through various instruments, namely UDHR, ICESCR, CRC, CEDAW and ICERD. Such pieces formulate the right to receive education and not to be discriminated against whilebeing educated from a universal perspective. The convergence of right to education and universalism is best understood through purpose-oriented policy. This depicts a universalpolicy of education, built on a number of elements. These have been propounded by the United Nations as availability, accessibility, acceptability and adaptability. These elements form the backbone of a universalist right to education and are necessary for the goal of universalism to be achieved. 

AI programs are at all times, and subject to prior internet connectivity, available to nearly everyone. AI systems, especially the future of continuous learning AI based on enhanced algorithms ensure adaptability allowing such systems to learn from new data at an unprecedented level. Additionally, the development of Adaptive AI provides greater agility for AI systems – providing them with the capability of continuously learning on new data, and swiftly adapting to contemporary circumstances in the real-world. Of particular importance is how AI has adapted itself for children with special needs – with personalized experiences. Adapting to specific needs of a student, AI ensures skill development and learning for those who are at a disadvantage in traditional learning. This evidence shows that AI programs are at par with the elements of availability and adaptability, however, the same cannot be said about the remaining elements for universalism i.e. accessibility and acceptability. 

Accessibility 

The element of accessibility as provided in the international framework intending for educational programmes is tri-dimensional in nature. The first facet of accessibility entails non-discrimination – education to be accessible to all, especially members of vulnerable groups. The second aspect of accessibility revolves around physical accessibility, ensuring that education is within safe and reasonably convenient geographic proximity or accessible through modern technology.  Economic accessibility is the third and last facet of the criterion of accessibility, necessitating the affordability of education for all.

Chat-bots used for educational purposes such as Chat-GPT, Duolingo, Mondly, and Londy AI operate with prior internet connectivity. Populations which lack resources to access internet connectivity cannot avail these platforms for any purpose, rendering it economicallyinaccessible to a great many people across the globe especially those below the poverty line. Usage of such models for educational purposes would put the well off at a greater advantage – allowing them to benefit from information quickly without having to utilize conventional methods of education. This sets groundwork for discrimination in the right to education, falling foul of accessibility as interpreted under international law. 

The inherent discrimination in the use of AI chat-bots is a reiteration of the endemic decadence present globally. When accessibility depends on prior internet connectivity, it becomes a privilege for certain social strata, leaving areas without the internet unable to benefit from these technological advancements. As a result, the society as a whole is unable to reap the benefits that accrue from such technological advancements. Generating differences even in the most basic of rights such as education, reduces globalization and stabs universalism to its core. Therefore, a holistic and accessible-to-all facility is necessary to counter the harm done by such AI models to the universality of the right to education.

Acceptability 

As AI primarily utilizes the data fed to it, it is prone to generating responses which are unacceptable to its users for a myriad of reasons. This unacceptability has especially been observed through the manifestation of bias and discrimination in AI responses. Since women and minorities are underrepresented in the available data, AI generated responses are prone to exhibit or perhaps exacerbate the existing prejudices against such vulnerable groups. Such biases have been highlighted through the AI systems employed in the healthcare sector.Besides this, notable AI-models such as Google’s online advertising system exhibited gender discrimination by displaying profitable jobs to male users more frequently than their female counterparts. Gender bias remains prevalent in AI-models for image generation as well, showing more men than women in specialized professions whereas, targeting of racial minorities persists at the hands of predictive policing tools which reinforce prejudicial views toward such communities. 

Bias infiltrates such algorithms through human bias (both explicit and implicit). Such bias is hardwired into humans, and consequently, the training data collected by humans is prone to containing these biases. Therefore, when AI systems process this data, they may inadvertently perpetuate, and in certain cases, exacerbate human biases, reflecting historical or social inequities. These flagrant errors embedded in internal databases forge inappropriateresponses, making AI-models irreconcilable with principles of non-discrimination. AI-chat bots used for education are also plagued by similar issues – the data stored in them is Euro-centric in nature and runs counter to the notion of universalism. Its continued use has refashioned pedagogy in a manner which confirms dominant narratives precluding universalism from education. 

Responses prepared by AI depict ideologies western in nature, omitting the role other cultures or societies have played, as evidenced by a piece prepared by Tamara N. Lewis Arredondo. In terms of human rights pedagogy, chat-bots have shown a tendency to omit occidentalism and colonialism, in the role of formulating human rights. Both concepts aided significantly in the formation of human rights, however, the failure of AI to keep them in loop illustrates the distortion of reality through ill-informed responses at least to the extent of human rights pedagogy. 

A history lesson on human rights violations from AI is distorted as well. Chat-GPT when asked of human rights violations gives a long list of states precluding the West and Europe glamorizing them as the Elysian Fields. When critically asked about this failure of omitting Western states, Chat-GPT accepted its fault and acknowledged a few violations which the Western powers have carried out. Following the confrontation that precluded West from its responses, Chat-GPT promised to ensure that Western violations will be included in the next responses. However, none was added in the later responses to prompts which raises several eyebrows as to the machine learning aspect of Chat-GPT as well.

Thus, without critical prompt techniques and making AI-bots criticise their own responses, AIwhen asked for general information will reinforce dominant views. Obtaining information for educational purposes from AI is unacceptable as it would brainwash users in accepting western oriented information. This would lead to more dissonance amongst different societies across the globe and would disseminate prejudicial views which run counter to universalist values. The same could be said with respect to the biases and discrimination emanating from other AI-platforms mentioned above. The absence of holistic information for educational purposes and inherent discrimination in AI-models make them unfit and unacceptable at least in their contemporary form for the purposes of education. In order for AI powered systems to be used in the realm of education, appropriate safeguards, such as ensuring use of diverse and representative data along with periodical human oversight, if not continuous, are essential steps that must be taken. 

Aggravation

Instead of rectifying the wrongs, the potential for harm has been exacerbated by the ‘Black box’ operation of AI systems. This entirely runs counter to how AI systems should be operating. Under the said operation, the internal mechanism of AI models remain opaque both to its users and developers. The secrecy under which these systems function take a toll on the forth-coming right to explainability i.e. right to obtain an explanation of the decision reached by an automated decision-making system. The use of AI systems, being marred by their black box operation, further stir a sense of uneasiness for their use in all realms including education. AI models used in various fields, such as law enforcement, have demonstrated that oversight is crucial, particularly when these models result in wrongful arrests and discrimination against minorities. Education delivered through methods prone to unjust results or findings raise concerns about its acceptability, challenging the concept of universal education. The lack of transparency in AI systems that may be used for the purposes of education is thus one of the grave issues which this technology has to overcome in order for it to be used efficiently for education.  

Way Forward?

AI, especially in the contemporary unregulated environment, remains a significant threat to the fundamentals of a society and to counter its impact, a way forward must be neatly identified. A ‘watch-dog’ or an international regulatory framework is a must for a sustainable way forward – the same has been expressed in the US Biden-Harris Executive Order on AI. Transparency, accountability, data privacy and non-discrimination should be the starting point of this regulatory framework. More importantly, the data fed in internal databases should be vetted, and be made as holistic as practically possible. Models that focus on education are to take into consideration how its curriculum can be made acceptable in nature, especially by moving away from solely Western dominant narratives. Furthermore, any attempts made domestically to safeguard issues posited by AI through legislation shall be more than welcome. It is under these frameworks that one may envision embracing innovation in a more secure, and universal manner; perhaps the only manner acceptable.

Conclusion 

In today’s digital age, the right to education remains insecure in particular, because of the threat posed by AI technologies to its universality. The current AI models not only reinforce a Euro-centric or Western-oriented standard through dissemination of one-sided and partial information, but also serve as a source of continued discrimination against already vulnerable communities. The black-box operation of AI systems further dents the right to education as these AI systems fall short of providing explanations for their decisions, reflecting negatively on accountability, transparency, and trust of such systems. There is a dire need to redress the contemporary AI-models that affect principles of universality and transparency through international and national regulatory frameworks. Proper vetting of the data fed to AI systems coupled with a human rights approach is another viable way of rectifying the problems posed to the right to education by AI systems. 

Disclaimer: Any and all opinions and views represented in this blog are personal and belong solely to the author(s) of the blog and do not represent the opinions or views of the Centre for Human Rights.