Introduction (Michelle)
Like many teachers of international law today, I was initially a little anxious about how to confront the challenge of AI in the classroom. Almost immediately, it was clear to me that striking an ostrich stance would not work. Joseph Weiler echoes these sentiments when he suggests that it ‘we would be remiss if in our design of courses (and consequently exams) we stick our heads in the sand and pretend that AI does not exist.’ Many of my students have been experimenting with AI’s imperfect search capabilities or its responses to essay questions. How could I devise an assessment exercise that developed their own critical thinking on the law while using AI? Was this even possible? In short, it was, and the assessment I devised was much more fruitful than I had anticipated. Here, in this post which is styled as a teacher and student response, I introduce the assessment I used last year before showcasing one student assignment by Katherine May. I conclude by reflecting on the lessons learnt from this experiment as a way to inspire variations for colleagues in the future.
Last year, I taught a 10-week Masters seminar called ‘Women’s Rights as Human Rights?’ to around 28 students. I retained the rather formulaic individual research essay as the main task, but I introduced a preliminary assignment based on ChatGPT. In week 3, we read and discussed the European Court of Human Rights decision, SAS v France (2014) in relation to questions around gender and cultural relativism. The case is a notorious example of a largely Orientalist approach to veiling, especially in relation to the wearing of the niqab. It considers claims that France’s full-face veil ban infringes Article 3 (inhumane treatment), Article 11 (the rights to freedom of association and discrimination in relation to this right) and Articles 8, 9, 10 (on the right to private life, freedom to manifest one’s religious beliefs and freedom of expression) as read together with Article 14 (relating to discrimination on specified grounds). The Grand Chamber acknowledged that French legislation banning the niqab presented the applicant with a dilemma: ‘either she complies with the ban and thus refrains from dressing in accordance with her approach to religion; or she refuses to comply and faces criminal sanctions’ (para. 110). The Court found that the ban imposed…can be regarded as proportionate to the aim pursued, namely the preservation of the conditions of “living together” as an element of the “protection of the rights and freedoms of others”…The impugned limitation can thus be regarded as “necessary in a democratic society”. This conclusion holds true with respect both to Article 8 of the Convention and to Article 9’ (157-158). It went on to hold that in addition there was no violation of Article 14.
This is a good case to generate class discussion. The facts are rather simple and the decision is starkly aligned with a particular interpretation of French democracy. This tends to elicit strong responses from students. Devoting class discussion time to the case first ensured that all students had a solid foundation before launching into the assignment itself.
I had expected a lot from the assignment – case analysis, an understanding of feminist judging, the ability to instruct ChatGPT as a group and critical reflective writing. Fortunately, these goals were all met and exceeded. First, I wanted the students to think about the practicalities and implications of feminist judging especially in terms of its intersectional methodological potential. From class discussion, it was already clear that the judgment was inadequate in approaching questions of gender and religious identity. Thus, I asked all students to begin the assignment by reading three academic articles on feminist judging by Charlesworth, Hodson and Hunter. This is a now well-established practice that sees feminist scholars re-write both domestic and international cases as a way to showcase the latent liberatory potential of the law. Second, all students were asked to watch a short online ChatGPT course. Third, I asked students to meet in groups and formulate collective prompts to re-write SAS as a feminist judgment. This part of the assignment was challenging. It required devising clear and effective prompts to generate the desired judgment. Some groups tried this more than once and were surprised by the different results. Fourth, students then discussed the AI-generated judgment in their group and the extent to which it resonated with their expectations of ‘feminist judging’. Fifth, I asked each student to write a 1000-word reflection on the exercise. No citations were necessary and the use of the first-person pronoun was encouraged. I listed these questions as a way to structure answers:
- How do you understand feminist judging? Was there agreement on this in the group?
- What is distinct about feminist judging in general and as embodied here?
- How can this be conveyed through AI?
- To what extent was your group in agreement about the process of prompt-writing?
- What did you learn from the group?
- How can this exercise help you read judgments differently?
I wanted to gauge how students had engaged with their use of AI and how this helped them to appreciate the specific dimensions of feminist judging and the production of legal knowledge.
Students were quite unsure about the assignment to begin with and each group had its own particular dynamic, but once students started playing around with ChatGPT, they had something to work with. Some groups demonstrated impressive creativity and flair, leading some to pen their own poems that were included in their reflections. My morning of marking was memorably enjoyable and full of pleasant surprises.
Here is one response that captures the reflexive potential from the assessment:
Individual Student Response (Katherine)
When the class was first briefed on this reflective assignment, I felt unmoored, as I was unclear what the specific outcome of the assignment was meant to be and what I was expected to produce. However, as I worked through the process, I came to realise that there is no correct or right output. The process of reflection and the analytical experience is the output and for me that experience became one of an examination of epistemologies: interrogating knowledge simulation tools (ChatGPT) and the extent to which the simulated knowledge it produces is equitably accessible and inherently male gendered, and a critical analysis of new theories and productions of knowledge (feminist judging).
Meeting as a group, we created a list of fourteen key words we felt were core principles of feminist judging. We then prompted ChatGPT to re-write SAS v France to include these concepts and received two very different responses. One group member, who was using her brother’s paid ChatGPT account, received a 500-word response providing detail on all fourteen key concepts and how these could be taken into consideration when re-writing SAS v France. The other group members, who all used free ChatGPT accounts, received nearly identical 100-word responses stating that ChatGPT did not have access to SAS v France and so could not re-write it. OpenAI states that ‘usage of ChatGPT is free’ and while access to a basic ChatGPT account is free, equitable access to the simulated knowledge it is capable of producing, is not.
This discovery prompted a number of questions: who created ChatGPT, who gets access to a higher quantity and quality of simulated knowledge, and is that simulated knowledge gender-neutral? While ChatGPT is presented as a gender-neutral programme, once you look more closely, Gross (439) argues that ‘Gender biases in ChatGPT have seemingly no end’. 74% of employees in the data and AI industry are men and so it is likely that predominantly male staff created, trained and now maintain ChatGPT (37). Men represent 66% of ChatGPT users and so are more likely to use the paid services and access a higher quantity and quality of simulated knowledge than women. Lastly, as ChatGPT is an LLM that develops iteratively, it will learn from and reproduce the input of the predominantly male user.
Given the male gendering of ChatGPT at nearly every juncture, is it a feminist programme and can it re-write SAS v France, which was created by the totemic male gendered judging panel, as a feminist legal judgement? The answer to the first question, ChatGPT informed me, is no, it is not feminist. However, when we asked ChatGPT to re-write SAS v France using our fourteen core principles, it did identify some key facts in the case and apply the principles, such as suggesting judges ‘focus on the experiences of Muslim women in France and how their rights and choices intersect with issues of gender, religion, and ethnicity’. Is this sufficient to be considered feminist judging? Some may believe so, however, for me this normative response does not ‘put bold ideas into practice’ as feminist legal judging aims to do (Charlesworth, 480).
This then led me to consider the type of knowledge that we were seeking to produce – feminist judgements – and to probe these theories and practices. I went back to the fourteen key words and compared these to the principles in the three required readings. Of our fourteen principles only five of them were explicitly referenced in the three articles. I was surprised to see that the articles did not explicitly identify empowerment as a key principle. For me, the raison d’être of feminist judging is to interpret the law in a way which amplifies the voice and agency of women and in doing so empowers the individual and the community through jurisprudential precedents. I was, however, more surprised by the absence of ‘de-colonial and post-colonial approaches’ and ‘intersectionality’ as key principles. For the group, these were crucial to achieving the transformative change which feminist legal judging purports to aspire to. The articles had referenced inclusivity, but they had done so in a general way, such as ‘including women’, which did not capture the concept of intersectionality (Hunter, 35). For me, the absence of de-colonial and post-colonial approaches and intersectionality as key principles of feminist legal judging is a serious failing. The status-quo is therefore only challenged from the Western context and judges are not challenged to de-construct racialised and oppressive structures or address the specific contexts and challenges in which each woman exists.
This also made me consider what the group had not included in our list of key concepts and why. We had not included the expectation that the feminist judge should be female, make ‘feminist choices’ or ‘commit herself to full-time feminism’ (Hodson, 928). As a group we did not discuss why we did not include certain terms, which on reflection was remiss of us, but for me, the reason why I had not proposed these terms is because they are so abstract that they become meaningless and empty Western scholarly proselytising. Why does a feminist judge need to be female? Feminism and gender equality benefits all genders and cannot be achieved in a vacuum. What are feminist choices? My feminist choices may be very different to someone else’s, but that does not mean that one is more legitimate than the other. What is full time feminism? Is it a 40-hour working week, or is it every hour of every day? Not all people have the resources or agency to make feminist choices at all times or to commit themselves to full time feminism. Some aspects of the articles therefore feel like they fail to take their own advice, and rather than taking prefigurative and transformative approaches, which produce new knowledge outside the normative framework, they instead reproduce and reinforce normative tropes, such as the disconnected Western scholar.
At the end of this assignment, I do not think that the group succeeded in using ChatGPT to re-write SAS v France as a feminist legal judgement. However, that endeavour now feels quite irrelevant. The reason being, I have gained so much more through this process of reflection and interrogation of knowledge production than I would have done, had ChatGPT simply produced the perfect feminist legal judgement.
Conclusions (Michelle)
A number of responses resonated with these insights from Katherine. The exercise empowered students to grapple with the imperfections of AI as feminist judging as a way to think through their responsibility in generating knowledge. The students gained significant confidence not only in using AI critically, but in thinking about the contingency and biases of human-generated legal thinking as well. I will be using similar assessments in the future to learn from my students through ‘dialogical teaching’ (Freire), as an ongoing form of knowledge co-production straddling both human and non-human capacities.
