As the practice of red-teaming for Responsible AI (RAI) evolves into a new digital profession, what are the emerging workforce challenges, and how can we support them? (challenge-code MSR-C6)
Generative AI technologies present both unprecedented capabilities and potential risks. As part of deployment safety and Responsible AI (RAI) processes, red-teaming has emerged as a key component of the responsible AI process, pivotal to vetting generative AI for public deployment.
RAI red-teaming, like content moderation and data curation and labelling, are novel forms of what has been called “data enrichment work.” It is needed to reduce the risk of generating harmful content and to improve the social robustness and safety of AI systems. And, like other forms of content moderation, red-teaming depends on people probing AI models for harmful outputs and then developing mitigation strategies to avoid such outputs.
This adversarial approach, combined with more traditional content moderation, requires exposing red-teamers to potentially disturbing content that would, otherwise, circulate to the broader public, post-deployment. However, the occupational health and psychological impact of such critical data enrichment work, as well as the tools and processes necessary to support that work, remains largely unexplored.
Microsoft is committed to investigating this process and the effects it has on people as part of our efforts to understand the psychological impact of RAI red-teaming. We are interested in assessing the diverse occupational health and psychological effects across content activities and identifying workplace coping strategies and support structures.
We are actively developing comprehensive guidance on managing the human infrastructure, from recruitment to psychological support for all data enrichment workers. And we are investigating how AI tools can enhance data enrichment work, minimize potentially harmful exposure, and understanding the broader implications of RAI red-teaming as a new digital professional.
Our research team consists of experts in the field of HCI with experience in conducting research in interdisciplinary topics such as mental health, social sciences, and cognitive psychology. The research team has mixed methods background from deeply qualitative, contextual, and ethnographic work to highly computational work involving machine learning, data science, and system development and deployment.
We welcome proposals that:
- complement the skill set of our team;
- are interdisciplinary in nature: we welcome both researchers based in core arts and humanities subject areas as well as researchers with expertise in other highly relevant disciplines, especially clinical psychology and organisational sociology, who can employ various methods appropriate to the challenge needs;
- draw on relevant expertise – examples include secondary trauma, occupational wellbeing, harmful content exposure, occupational wellbeing, data enrichment work, remote/hybrid work and organizational support structures;
- consider evidence-based interventions or ideas around recruitment, onboarding, training, and maintaining the human infrastructure for responsible AI; and
- benefit from a flexible and collaborative approach.
How it will work
- If the application is successful, MSR will work with the fellow to refine the project plan, agreeing on specific goals and outcomes as well as a timeline for shared collaboration milestones and a cadence for meetings virtually and in-person, as appropriate.
- MSR will support the fellow in terms of onboarding and providing an MSR research contact and engaging in regular meetings.
- MSR is set up for hybrid working and we extend a warm welcome to the BRAID fellows to visit MSR Redmond and/or New England labs, should they find themselves interested and available. Rest assured, we are committed to offering the utmost support to ensure a productive and rewarding experience.
- We expect the fellow to factor into their budget proposal travel, accommodation and subsistence costs, and any specific research costs they envision.