Incorporating ChatGPT into Human Rights Pedagogy and Research Practices

[Tamara N. Lewis Arredondo is a Senior Lecturer and researcher on human rights for the Global Citizenship research group at The Hague University of Applied Sciences, The Netherlands]

What can ChatGPT teach you, me and our students about human rights?  As with all new technology, generative AI has impacted the academy, influencing how we teach, assess and even how we acquire knowledge in all fields, including human rights. Now that the initial hype surrounding the release of ChatGPT is beginning to settle, we can explore ways generative artificial intelligence (AI) can be ethically employed in human rights law education. That exploration should be informed by a comprehensive understanding of what perspectives, viewpoints and narratives the tool will return after a prompt.

My Method for Testing ChatGPT’s Understanding of Human Rights

In my own academic practices, I counter dominant narratives surrounding the historical origins of human rights. That includes a critique of the iconic positioning of the Universal Declaration of Human Rights through queering its language and broadening its focus to include other-than-human lifeforms. My work demonstrates that the very instruments designed to protect can and do exclude the beings that are most in need of their protection. 

In this piece, I recount my experiences prompting ChatGPT about human rights. Firstly, I present the Gartner Hype cycle as a way to frame both early responses to, and the pathway of, emerging technology. After describing the features of Open AI’s ChatGPT, I introduce three prompts regarding the history and origins of human rights into ChatGPT’s free and paid versions to perspectives and viewpoints favored by the tool. Then, I introduce other prompts designed to test ChatGPT’s ability to find critical or differing perspectives regarding the history and origins of human rights. Finally, I end the discussion with my findings about ChatGPT and how it can be used to assist in pluriversal understandings of human rights and their origins.

My experiment is guided by my hypothesis that, just as search engines like Google tend to reproduce dominant societal (US-based) narratives in their content ranking of websites, ChatGPT will return information about human rights that is steeped in the dominant narrative of the field. For me, the dominant narrative in human rights is that story that human rights originated in the philosophical and political thinking of Greek civilization, the Enlightenment, John Locke, the Declaration of Independence of the US and France’s Declaration of the Rights of Man, and that the Second World War created the moral urgency to create legal instruments to protect the rights of all people through the newly-founded United Nations.

Understanding How New Technologies Behave

The Gartner Hype Cycle is a commercial model to understand how expectations regarding newly innovated technology evolves over time. It holds that a successful innovation will move through five stages over a period of three to five years, starting with a strong peak as the release of the tool generates excitement through media coverage and hype around the product.  That high point is immediately followed by a decline known as the “trough of disillusionment” when the weaknesses of the product are discovered and some users abandon the market. Eventually, if the technology is to survive, it should reach a “plateau of productivity” where it can continue existing over time. 

This cycle is useful for understanding ChatGPT. If, as researchers and lecturers, we can use the technology proactively and ethically, ChatGPT attains the productive phase and can assist us over time. 

How LLMs work

ChatGPT, like other large language models (LLMs), is a predictive text generator “trained”on very large data sets, including books, articles, and web pages. After “learning” that data, the model can respond to prompts from humans for information in natural words. 

While ChatGPT will answer in human language, it is a tool. Any answers it generates will only be as good as the information it has learned. Equally important, the prompt quality greatly influences the relevance of the response. ChatGPT can give false or misleading information. This is a phenomenon referred to as “hallucination.” Accordingly, users need basic knowledge about a topic to recognize when the tool has produced a false or misleading response.

Finally, as with many technological solutions, OpenAI is a business solution. Though the underlying technology that runs ChatGPT is open source and other companies and start-up ventures use and develop it, Open AI has a tiered platform. The free version (Chat 3.5) varies significantly from the paid version (Chat 4.0) and the results for prompts in the paid version are based on quick searches on the Internet search engine, Bing, and include links to sources. No such links are provided in Chat 3.5.

Exploration: Initial Prompt and Responses

I began with a very basic prompt (Prompt 1) in both the free and paid versions: what are the origins and history of human rights? 

The results of Chat 3.5 included eight points: Ancient Civilizations, Greco-Roman influence, the Magna Carta, the Enlightenment, The American and French Revolutions, 19/20th Century events and the Universal Declaration of Human Rights.

The Chat 4.0 yielded six points. The ancient and religious foundations, Philosophical influences, The Enlightenment, the 19^th and 20^th Centuries, Post World War II and the Modern Era. 

Here, the responses confirmed my original hypothesis. Interestingly, both results reference non-Western origins. Version 3.5 points to Dharma from ancient India’s civilization. Both versions acknowledge that human rights are expanding to encompass environmental rights and LGBTQ+. Version 4.0 highlights religious teachings from Islam, Hinduism and Buddhism that “echo concepts of human dignity.” It further identifies cultural relativism political interests and enforcement issues as “challenges and critiques.”

Critique of the Responses

The responses given in both versions closely resemble what my research students and I have observed in our study of textbooks and syllabi of introductory human rights courses, namely, that the origins and history of human rights are largely associated with Western historical events emanating from the Enlightenment and the American and French Revolutions. The content and substance of the ChatGPT responses reinforce the dominant narrative, planting the seeds of human rights in a linear historical and philosophical progression of Western Civilization. This tendency to link human rights to Western ideologies has been widely critiqued by TWAIL scholars and by decoloniality scholarship emanating from Latin American theorists.

While both of the ChatGPT responses briefly reference the abolition of slavery, listing such an event on its own without referring to the rise of the Trans Atlantic slave trade provides a misleading context for the origins of human rights. It is almost as if the abolition is a precursor to modern human rights, but the products and drivers of the slave trade are not pertinent to the eventual creation of the human rights regime.

The ChatGPT response, like the human rights textbooks and syllabi omit two very relevant phenomena that originated in the West and heavily influence the development of human rights: colonization and capitalism. Any pedagogy looking at the origins and history of human rights must study the lasting impact of colonial empires and extractive capital on our existing human rights regimes.

Further Prompting to Broaden the Scope 

I made two attempts to broaden or deepen the description of the origins and history. Firstly, I prompted as follows (Prompt 2):  Summarize scholarly critiques to the origins and history of human rights that you produced above. Where relevant, give the names of the scholars and the school of thought as well as links to the sources.  The results from Chat 4.0 were very narrow and only focused on Samuel Moyn.  The results for 3.5 were more general with no mention of scholars. The Chat 3.5 tool reminded that it “cannot provide specific sources or links to scholarly articles due to [its] knowledge cutoff date and the limitations of [its] browsing capabilities.”    

I ended with a third, very specific prompt (Prompt 3), asking “How would a TWAIL scholar define the origins and history of human rights?” The results included references to three TWAIL scholars with links to their articles. Chat 3.5 remained general and no scholars were specified.

Conclusions About Future Human Rights Pedagogy and Research Practices

ChatGPT versions 3.5 and 4.0 have been trained on materials that, when asked to explain the origins and history of human rights, reinforce the dominant narrative about human rights; one that closely resembles that which is outlined in introductory human rights textbooks; namely that of a progressive path starting with ancient texts, including the Code of Hammurabi, and traveling through the Greek and Roman empires, the Enlightenment, the American and French Revolution, the abolition of slavery and the Second World War to dish up the numerous treaties and conventions that exist today. Only after being prompted to critique its own response, will ChatGPT explore other perspectives. 

While it is possible to get views from non-dominant perspectives, it seems that familiarity with the schools of thought or the scholars who offer such knowledge is crucial for adequate prompting for a comprehensive response from ChatGPT. 

Two important lessons emerge for human rights law teaching practices. Firstly, teach students to use reflective and critical prompt techniques asking ChatGPT to find critiques to its own responses. Additionally, other follow up critiques should specify the ideologies, methods, scholars, or schools of thought to explore, like I did in the third prompt.

Simple prompts asking for general information yield responses that reinforce dominant views on topics regarding human rights. Therefore, this tool needs very specific prompts if one wants to go beyond common understandings in the field. In academia, this means guidance from more experienced and knowledgeable persons who are familiar with alternate viewpoints and perspectives in the field. Using one’s expertise, ChatGPT can be prompted to summarize the viewpoints of TWAIL scholars or other critical viewpoints, such as ecofeminism, eco socialism, critical legal theorists and critical race theorists. Furthermore, the prompts within Chat 4.0 should ask for the names of scholars and links to the sources for verification and further research. 

Secondly, to bring ourselves to the plateau of productivity (or as I like to think of it, the plateau of proactivity), we should train our own bots, which will be built in Chat 4.0 using Open AI’s technology and incorporating the scholarship we find crucial  – TWAIL, Critical Legal Theory and Decoloniality. This project will require compiling a repository of PDFs of important scholarship and careful instructions to the bot regarding what to prioritize when considering texts. The results will be promising for developing tools that customize and refine the technology to produce more pluriversal tools for human rights pedagogy. Additionally, because those bots can also tell users what is in its database, they will be more transparent to users.

Within a bot building project, I foresee a lot of promise for the future of human rights pedagogy and research. The ability to build bots offers an excellent opportunity to create custom instruction so that students can become well versed in certain aspects of human rights. It can be a collaborative and long term effort where content or data is amassed and fed to the bot, thereby broadening the scope of the responses to prompts. These bots can be used by students who can prompt it for deepening understanding of the critiques to dominant narratives surrounding the human rights field. It could also be a good way to use “glocal” knowledge thereby centering knowledge from the Global South or indigenous communities. 

Conclusion

This exploration with ChatGPT underscores two vital lessons for human rights law education. First, the importance of reflective and critical prompting techniques that challenge it to critique its responses. Second, the potential of customizing AI tools like ChatGPT, incorporating diverse scholarly perspectives to foster a more inclusive and comprehensive understanding of human rights. It also shows the promise of using collaborative approaches to build tools that help create pluriversal approaches to the study of human rights law.