RAI theory, practice, and critique

We critically analyse current methods and practices of Responsible AI, and develop new theories and frameworks for better practice.

Publications

1. A. Ghoshal, A. Dutta, R. Abu-Salma, S. Modgil, and M. Brandao, “Responsible AI as Hegemony: Epistemic Exclusion, Technosolutionist Conformity, and Performative Compliance Among AI Practitioners in India,” in AAAI/ACM Conference on Artificial Intelligence, Ethics, and Society (AIES), 2026. [Abstract] [PDF]

   This paper investigates how AI practitioners in India perceive and experience Responsible AI (RAI) through mixed-methods qualitative research with participants from industry and academia. We argue that RAI practice in India operates as a doubly hegemonic system in which global Western frameworks that privilege technical and procedural conceptions of responsibility converge with local caste-gender hierarchies to determine what counts as legitimate ethical knowledge. Drawing on interviews with 18 practitioners, we show how this double hegemony operates through three interlocking mechanisms. Counter-hegemonic perspectives, particularly corrective justice framings articulated by caste-marginalised practitioners, are shadowed through epistemic silencing. Surviving critique is channelled into technosolutionist forms that leave organisational power intact. The system is sustained through performative compliance, where organisations display RAI commitments while protecting institutional priorities from ethical interruption. These mechanisms produced conditions in which practitioners regularly recognised harm but lacked the authority, protection, or collective support to act on that recognition. We propose that countering RAI hegemony requires collective and enforceable infrastructures, including Corporate Social Responsibility (CSR)-embedded ethics residencies, anonymous reporting mechanisms, and vernacular harm taxonomies grounded in community knowledge rather than imported risk categories.

2. A. Tolefat, M. Nwachukwu, and M. Brandao, “Analysing Demographic Representation in Wikidata: A Visual Analytics Approach,” in Wikidata’26: Wikidata Workshop at ISWC 2026, 2026. [Abstract] [Code]

   Wikidata is one of the largest open knowledge graphs, which contains data on real people’s demographic data at scale. This data is also known to be unevenly covered across gender and geography, a problem with important downstream effects. Identifying and analysing such gaps is currently challenging in practice, however, especially for non-expert users and contributors. This paper presents a visual analytics system for exploring and investigating demographic representation gaps, bias and incompleteness in Wikidata biographies. The system gathers country-level demographic data across five demographic dimensions: gender, age, language, ethnic group, and occupation, with gender additionally analysed over time. These are visualised in an interactive dashboard that combines a global country overview with per-country views. The dashboard enables non-expert users to examine demographic patterns, identify representation gaps within Wikidata, and plan interventions towards more equitable representation of content. We used a controlled between-subjects study with forty (n=40) participants to evaluate the dashboard in representation-gap investigation tasks, showing fast task completion times, higher task accuracy and confidence than a tabular data baseline, and a System Usability Scale score of 84.25. Thus we demonstrate that incomplete knowledge graph records can be transformed into interpretable representation profiles that enable non-expert users to identify representation gaps and plan interventions towards more equitable content. Finally, we report on the current state of demographic representation in Wikidata, as of 2026, namely in terms of age distribution and the countries, age ranges and occupation sectors with highest (over 80%) gender gap.

3. A. Ghoshal et al., “Position: Responsible AI for AI companions must actively combat violence toward intimate partners,” in International Conference on Machine Learning (ICML) Position Paper Track, 2026. [Abstract] [PDF]

   AI companions differ from earlier interactive technologies by creating sustained relational environments through anthropomorphism, emotional engagement, and continuous validation. This position paper argues that Responsible AI for AI companions must actively combat violence toward intimate partners, including those who may never directly interact with these systems but may nonetheless experience the consequences of users whose behaviors have been shaped through prolonged engagement with them. We examine how these systems can create conditions in which users rehearse violent, coercive, or abusive behaviors without encountering meaningful resistance, and we identify structural gaps in existing safety approaches that focus primarily on protecting direct users. Drawing on research on intimate partner violence (IPV), coercive control, and technology-facilitated abuse, we propose three intervention pathways: involving IPV survivors in red-teaming and benchmark development; implementing behavioral monitoring with graduated enforcement mechanisms; and reorienting AI safety research toward granular harm taxonomies capable of detecting longitudinal patterns of violence across extended interactions. Together, these recommendations broaden the scope of AI safety by centering the security of non-users alongside the well-being of users.

4. A. Ghoshal, M. Brandao, R. Abu-Salma, and S. Modgil, “Embodied AI at the Margins: Postcolonial Ethics for Intelligent Robotic Systems,” in AAAI/ACM Conference on Artificial Intelligence, Ethics, and Society (AIES), 2025. [Abstract] [DOI] [PDF]

   As artificial intelligence (AI)-powered robots increasingly permeate global societies, critical questions emerge about their ethical governance in diverse cultural contexts. This paper interrogates the adequacy of dominant roboethics frameworks when applied to Global South environments, where unique sociotechnical landscapes demand a reevaluation of Western-centric ethical assumptions. Through thematic analysis of seven major ethical standards for AI and robotics, we uncover systemic limitations that present challenges in non-Western contexts - such as assumptions about standardized testing infrastructures, individualistic notions of autonomy, and universalized ethical principles. The uncritical adoption of these frameworks risks reproducing colonial power dynamics in which technological authority flows from centers of AI production rather than from the communities most affected by deployment. Instead of replacing existing frameworks entirely, we propose augmenting them through four complementary ethical dimensions developed through a postcolonial lens: epistemic non-imposition, onto-contextual consistency, agentic boundaries, and embodied spatial justice. These principles provide conceptual scaffolding for technological governance that respects indigenous knowledge systems, preserves cultural coherence, accounts for communal decision structures, and enhances substantive capabilities for Global South communities. The paper demonstrates practical implementation pathways for these principles across technological life cycles, offering actionable guidance for dataset curation, task design, and deployment protocols that mitigate power asymmetries in cross-cultural robotics implementation. This approach moves beyond surface-level adaptation to re-conceptualize how robotic systems may ethically function within the complex social ecologies of the Global South while fostering genuine technological sovereignty.

5. A. Ghoshal, M. Brandao, and R. Abu-Salma, “Value Alignment in the Global South: A Multidimensional Approach to Norm Elicitation in Indian Contexts,” in ICLR 2025 Workshop on Bidirectional Human-AI Alignment (BiAlign), 2025. [Abstract] [PDF]

   This paper addresses critical gaps in artificial intelligence (AI) value alignment research concerning historically marginalized communities in the Global South, with a specific focus on Dalits and Adivasis in India. We propose a multidimensional approach that integrates B.R. Ambedkar’s and Amartya Sen’s theoretical frameworks for social justice with Clifford Geertz’s thick description methodology to develop context-sensitive norm elicitation processes. By examining how deeply entrenched sociopolitical hierarchies influence these communities’ agency to process and express information about their own values, we demonstrate that conventional approaches to value alignment inadequately address the unique challenges faced by these communities. Our framework emphasizes the role of Indian AI missions in creating culturally relevant scenarios for norm elicitation, ensuring meaningful participation of marginalized communities in AI alignment processes. This approach not only advances the discourse on inclusive AI development, but also provides practical strategies for implementing value alignment methodologies that acknowledge and address historical power dynamics.

6. M. Brandao, M. Mansouri, and M. Magnusson, “Editorial: Responsible Robotics,” Frontiers in Robotics and AI, vol. 9, Jun. 2022. [DOI]
7. M. Brandao, “Normative roboticists: the visions and values of technical robotics papers,” in IEEE International Conference on Robot and Human Interactive Communication (RO-MAN), 2021, pp. 671–677. [Abstract] [DOI] [PDF]

   Visions have an important role in guiding and legitimizing technical research, as well as contributing to expectations of the general public towards technologies. In this paper we analyze technical robotics papers published between 1998 and 2019 to identify themes, trends and issues with the visions and values promoted by robotics research. In particular, we identify the themes of robotics visions and implicitly normative visions; and we quantify the relative presence of a variety of values and applications within technical papers. We conclude with a discussion of the language of robotics visions, marginalized visions and values, and possible paths forward for the robotics community to better align practice with societal interest. We also discuss implications and future work suggestions for Responsible Robotics and HRI research.

8. E. T. Williams et al., “Begin with the human: Designing for safety and trustworthiness in cyber-physical systems,” in Human-machine shared contexts, Academic Press, 2020, pp. 341–357.
9. C. Bentley et al., “Including women in AI-enabled smart cities: Developing gender-inclusive AI policy and practice in the Asia-Pacific region,” in AI for Social Good, APRU, 2020.
10. E. Nabavi, K. A. Daniell, E. T. Williams, and C. M. Bentley, “AI for sustainability: a changing landscape,” Artificial intelligence—for better or worse. Future Leaders, pp. 157–176, 2019.
11. E. T. Williams, C. Bentley, K. A. Daniell, N. Derwort, K. Leins, and E. Nabavi, “Complexity is not new: How our own technological history can teach us about AI,” in Artificial Intelligence: For Better or Worse, Future Leaders, 2019.