Bridging the Gap: Reflections from teaching translational data and AI ethics

James Garforth, Benedetta Catanzariti, Meenakshi Mani

Data and AI ethics initiatives and training programmes have become widely popular across sectors, although many have noted how the knowledge and tools gained through these efforts are difficult to “translate” into practice. While data and AI ethics principles and guidelines abound, data practitioners see these as too abstract and often ineffective within the practical contexts of technology development. Conscious of these concerns, we developed a curriculum focused on building “translational ethics” skills across disciplines, as part of the 2023/2024 EFI MSc programme ‘Data and Artificial Intelligence Ethics’. In this blog post, we reflect on our experience collaborating across social science and computer science to design and teach the new EFI “Translational Data and AI Ethics” intensive course. The course, which took place in February 2024, catered to students and practitioners with diverse backgrounds, including computer and data science, business, humanities, and social science.

The many meanings of ‘translation’

Translational ethics is a concept first originating in medical ethics, where ‘translation’ typically refers to the work required to bridge the gap between laboratory research and real-world clinical practice. In this sense, translational ethics involves the development of ethically justified approaches to the movement between the different stages in the production of medical knowledge: fundamental research, clinical trials, dissemination and, finally, the implementation of new practices and guidelines at a policy level (Bærøe 2014; Hostiuc et al. 2016). Translational ethics thus requires engagement with both theory and practice, and a commitment to interdisciplinary collaboration between various forms of knowledge and evidence to produce safe and effective pathways and treatments.

Drawing on these insights, the nascent field of ‘Translational Ethical AI’ (Borg 2022) proposes a similar approach to AI ethics, where ethical judgement is seen as integral to all phases of technology development, from foundational research to real-world AI deployment. Within this approach, discussions around technical choices cannot be separated from considerations around social and political impact, environmental sustainability, and accessibility. Rather than a standalone practice, ethics becomes here a form of knowledge essential to develop both ethically and technically sound systems (Goetze 2023).

An important point of consideration is that common high-level ethics approaches seem to assume equal and homogeneous agency and accountability within AI supply chains, in contrast with the often fragmented and modular reality of AI and data practice. Our objective in setting up the course was to break down these barriers and reconnect social and ethical understanding with the contexts of technology development.

We focused on unpacking the values and practices – disciplinary, institutional, and educational – that make up the social and organisational settings of technology development. Throughout the course, we explored the different contexts in which AI artifacts and practices are conceived (research), funded (investment), taught (education), and contested (labour organisation). By investigating the norms and values central to these different contexts, we aimed to develop tools that can help build a common language to translate ethical considerations into practice across disciplinary boundaries.

Mapping data and AI contexts

The course was divided into three sections, building on each other. We started by introducing our students to qualitative research investigating the social and organisational actors who form the data and AI industry: what groups tend to be more or less represented in the field, what values are typically promoted or reinforced within this industry, and what forms of knowledge practitioners inherit from traditional computer science training programmes, along with disciplinary norms and practices.

Next, we looked at the contexts and environments in which tech work is typically performed: what tends to get prioritised in research practices, how software development is usually carried out, and how organisational settings – from start-ups through to large corporations – can shape the ethical attention of their workers. These parts of the course sought to provide students with research-supported understanding of the current data and AI landscape: what social, organisational, and disciplinary norms shape technology development and its social impact.

In the final section of the course, we gave students practical examples of translational approaches to data and AI ethics, through a range of presentations and discussions with researchers and practitioners actively engaged in what we see as translational work. These guests were chosen from across sectors and backgrounds to represent a wide variety of practices, from education and worker organising, through small research projects, up to policy and venture capital investment, including a reflection on the potential pitfalls and challenges of these various approaches.

To help ground everything we had taught them, we capped off the course with a practical activity where groups of Translational Data and AI Ethics students were paired up with undergraduate students from the School of Informatics part way through the development of start-up style projects. This provided both sets of students with a rare opportunity to develop a shared language, and to practice the kinds of interdisciplinary conversations we hope to make more commonplace.

What did we learn?

The student response to the course was exceptional from the start, with meaningful engagement from everyone, leading to guest sessions often overrunning with questions and reflections. Students expressed excitement about the curation of readings provided (which included both ethnographic studies of, and technical literature on, AI and data practice) as well as the practical – yet theoretically-challenging – nature of the course activities. Furthermore, after the completion of the course, several students chose to revisit their dissertation project ideas to include more elements of translational ethics; for example, by investigating how different software development processes (waterfall vs agile) might shape developers’ societal and ethical considerations.

By creating space for rich and thoughtful conversation with students, this course gave us an opportunity to reflect on some of the broader challenges to the implementation of translational ethics teaching programmes within more traditional computer and data science degrees. For example, many students with an informatics background reported that, while they could now appreciate the importance of seeing technical practice as a social practice, they were all taught to code and build computer systems without this context. As already observed by others (Sarder and Fiesler 2022; Darling-Wolf and Patitsas 2024), even within institutions and programmes that do offer ethical training, technical skills are still taught as if entirely abstract and stripped of their social and ethical significance.

Some students, conversely, hoped that the course would include technical methods to implement ethics in code. We noted however that, while fairness metrics can serve as helpful tools to measure algorithmic bias and discrimination, these alone are not sufficient to mitigate the potential impact of data-driven practices (Corbett-Davies et al. 2023). ‘Representative’ data and ‘accurate’ predictions can still cause harm when AI systems are used in contexts that disproportionately affect certain groups over others, such as surveillance and policing (Eubanks 2014; Browne 2015). While some calls for translational ethics emphasise the need to understand how to “implement” ethics in code, our approach aligns with scholars who have pointed out how this narrower conceptualisation and operationalising of translation runs into the pitfalls of technosolutionism that give rise to ethical issues in the first place. Translational ethics, then, requires that technologists go beyond technical and statistical fairness, and situate technical approaches to AI and data ethics within the organisational, as well as broader social and political contexts of technology development and use.

Towards a translational ethics agenda

The course has also presented an opportunity to reflect on some of the wider institutional and cultural challenges to the implementation of a translational ethics teaching agenda. First, translational ethics requires sufficient engagement with both qualitative and quantitative forms of knowledge, contrary to assumptions typical within engineering fields over the epistemological superiority of “hard science” over “soft” knowledge (Raji, Scheuerman, and Amironesei 2021). Within computer science degree programmes, this would require a reconfiguration of traditional curricula currently positing ethics and social science-oriented modules as optional or peripheral to “core” engineering knowledge (Downey 2021); a position that is often mirrored in many ways by the tech industry (Darling-Wolf and Patitsas 2024). Secondly, students need to be taught to understand and identify AI ethics as contextual and specific, grounded in issues of social justice (Munn 2022; Amugongo et al. 2023). It is also crucial to ensure that translational ethics doesn’t merely focus on incorporating metrics, unreflectively, into technology. Finally, for all of this to be supported, further research needs to be done into what does or doesn't work in current translational contexts, so that we know which approaches are most likely to be successful. This could require a fundamental shift in the funding landscape, which is typically siloed, making sources of funding for truly transdisciplinary work hard to find.

Into the future

We will continue to run this course again next year, and have a new cohort of amazing students from varied backgrounds ready to participate and give us additional feedback. Based on this past year’s experience, it is clear that engaging more meaningfully with the norms and processes involved in standard software development cycles would be beneficial to students without a technical background.

We are also keen to contribute further to the growing body of research literature which led to the design of the course. In this regard, more research-led pedagogical scholarship is needed to truly equip students with the translational tools necessary for the creation of more socially-just computing cultures. This would include research into what makes for a successful translational intervention, what we can learn from translational divides between disciplinary norms and practices, as well as what hidden factors in traditional computer science education might shape students’ social and ethical knowledge.

About the contributors:

Dr James Garforth (CTMF Senior Research Affiliate) teaches ethics, social responsibility and teamwork to undergraduate students in the School of Informatics, and supervises projects to develop tools and practices supportive of responsible development.

Dr Benedetta Catanzariti (CTMF Postdoctoral Affiliate) is a British Academy Postdoctoral Fellow in Science, Technology and Innovation Studies. Her work explores the social, ethical, and political dimensions of data-driven technologies, with a focus on machine learning and its related data practices.

Meenakshi Mani (CTMF PhD Fellow) is an interdisciplinary researcher with experience in the fields of computer science and education who is critically examining how EdTech engineers conceptualise and construct AI education technologies.

Bibliography

Amugongo, Lameck Mbangula, Alexander Kriebitz, Auxane Boch, and Christoph Lütge. 2023. ‘Operationalising AI Ethics through the Agile Software Development Lifecycle: A Case Study of AI-Enabled Mobile Health Applications’. AI and Ethics, August. https://doi.org/10.1007/s43681-023-00331-3.
Bærøe, Kristine. 2014. ‘Translational Ethics: An Analytical Framework of Translational Movements between Theory and Practice and a Sketch of a Comprehensive Approach’. BMC Medical Ethics 15 (1): 71. https://doi.org/10.1186/1472-6939-15-71.
Borg, Jana. 2022. ‘The AI Field Needs Translational Ethical AI Research’. AI Magazine 43 (3): 294–307. https://doi.org/10.1002/aaai.12062.
Browne, Simone. 2015. Dark Matters: On the Surveillance of Blackness. Duke University Press.
Corbett-Davies, Sam, Johann D. Gaebler, Hamed Nilforoshan, Ravi Shroff, and Sharad Goel. 2023. ‘The Measure and Mismeasure of Fairness’. arXiv. https://doi.org/10.48550/arXiv.1808.00023.
Darling-Wolf, Hana, and Elizabeth Patitsas. 2024. ‘“Not My Priority:” Ethics and the Boundaries of Computer Science Identities in Undergraduate CS Education’. Proc. ACM Hum.-Comput. Interact. 8 (CSCW1): 174:1-174:28. https://doi.org/10.1145/3641013.
Downey, Gary Lee. 2021. ‘Critical Participation: Inflecting Dominant Knowledge Practices through STS’. In Making & Doing. Activating STS through Knolwedge Expression and Travel, edited by Gary Lee Downey and Teun Zuiderent-Jerak. The MIT Press. https://direct.mit.edu/books/oa-edited-volume/5153/chapter/3403234/Critical-Participation-Inflecting-Dominant.
Eubanks, Virginia. 2014. ‘Want to Predict the Future of Surveillance? Ask Poor Communities.’ The American Prospect. 15 January 2014. https://prospect.org/api/content/36656b9e-c446-5205-9257-0120f64aabdb/.
Goetze, Trystan S. 2023. ‘Integrating Ethics into Computer Science Education: Multi-, Inter-, and Transdisciplinary Approaches’. In Proceedings of the 54th ACM Technical Symposium on Computer Science Education V. 1, 645–51. SIGCSE 2023. New York, NY, USA: Association for Computing Machinery. https://doi.org/10.1145/3545945.3569792.
Hostiuc, Sorin, Alin Moldoveanu, Maria-Iuliana Dascălu, Runar Unnthorsson, Ómar I. Jóhannesson, and Ioan Marcus. 2016. ‘Translational Research—the Need of a New Bioethics Approach’. Journal of Translational Medicine 14 (1): 16. https://doi.org/10.1186/s12967-016-0773-4.
Munn, Luke. 2022. ‘The Uselessness of AI Ethics’. AI and Ethics, August. https://doi.org/10.1007/s43681-022-00209-w.
Raji, Inioluwa Deborah, Morgan Klaus Scheuerman, and Razvan Amironesei. 2021. ‘You Can’t Sit With Us: Exclusionary Pedagogy in AI Ethics Education’. In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, 515–25. FAccT ’21. New York, NY, USA: Association for Computing Machinery. https://doi.org/10.1145/3442188.3445914.
Sarder, Ella, and Casey Fiesler. 2022. ‘Entering the Techlash: Student Perspectives on Ethics in Tech Job Searches’. In Companion Publication of the 2022 Conference on Computer Supported Cooperative Work and Social Computing, 85–88. CSCW’22 Companion. New York, NY, USA: Association for Computing Machinery. https://doi.org/10.1145/3500868.3559446.