SRI graduate fellows convene workshop on designing responsible AI futures
Organized by SRI’s 2025–26 graduate fellows, Designing Responsible Futures examined how AI can be shaped through operational safety, community participation, worker-led governance, and democratic principles.
On April 29, 2026, the Schwartz Reisman Institute for Technology and Society’s 2025–26 cohort of graduate fellows hosted Designing Responsible Futures, a graduate workshop bringing together researchers, practitioners, and emerging scholars to examine how increasingly capable AI systems are reshaping responsibility, authority, and governance.
The half-day workshop moved from immediate questions of responsible AI development to broader issues of expertise, power, and long-term futures. Across two sessions and a graduate fellow poster session, speakers explored how AI systems are being built and deployed in practice, who has the authority to shape their development, and what kinds of governance structures are needed as AI becomes more deeply embedded in public, institutional, and professional life.
Building trustworthy AI systems
The first session, “Responsible Development: Building Trustworthy AI Systems,” focused on the practical and institutional conditions required to make AI safer, fairer, and more accountable. Moderated by SRI Graduate Fellows Mai Ali and Benjamin Cookson, the session examined responsible AI as a challenge that extends well beyond technical model design.
Opening the discussion, Shingai Manjengwa, Senior Director, Education and Development, Talent & Ecosystem at Mila, argued that responsible AI must be understood as a people and operations problem, not only a matter of principles or compliance. Drawing on her work teaching AI governance to policymakers and industry leaders, Manjengwa described how common tools such as ethics boards, impact assessments, and regulatory checklists remain important, but are not sufficient for AI systems that can operate autonomously, at scale, and at machine speed.
Using a case study of an enterprise AI security incident involving McKinsey’s internal AI platform Lilli, Manjengwa emphasized that the risk is not only unauthorized access to data, but the possibility that an attacker could alter what an AI system tells thousands of users.
“When we think about AI governance, let’s move from the softer side of governance to the hard operational data side of governance,” Manjengwa said. “We need it to be part of the operational culture. We need incident investigation infrastructure. We need shared accountability mechanisms, and we need continuous, not periodic, oversight.”
For Manjengwa, mature safety cultures in aviation, healthcare, and nuclear systems offer useful lessons for AI governance. In aviation, incidents trigger independent investigation, public learning, and operational changes designed to prevent future failures. AI, she argued, needs similar structures. “The governance for that must look different,” she said. “It is a people problem. It is an operational problem.”
Laura Rosella, professor at the University of Toronto’s Dalla Lana School of Public Health and a Canada Research Chair in Population Health Transformation and Analytics, brought the conversation into the health context, where AI carries both extraordinary promise and significant risk. Rosella described health systems facing growing pressures from capacity challenges, workforce burnout, medical complexity, and fragmented care, while also cautioning that AI tools can deepen inequities if they are not deployed carefully.
“If we’re going to direct AI somewhere to make it better, it should be here,” Rosella said, referring to healthcare. “It’s also one of the highest-risk areas. We have the most potential benefit and the most potential risks.”
Rosella’s presentation identified four tensions shaping responsible AI in health: the paradox of progress, the scale-up problem, the gap between human-centred design and real-world deployment, and governance across organizational boundaries. An AI tool may work well in a controlled pilot, she noted, but fail when scaled across different sites, workflows, populations, and data environments.
Responsible AI in health, Rosella argued, requires meaningful engagement with the people most affected by these systems. Drawing on a project deploying diabetes prediction models in Peel Region, she described an approach that began not with model performance, but with community and practitioner engagement.
“We did not start with the model,” Rosella said. “We started by bringing people together around the problem, then co-designing deployment with practitioners and community partners.” Responsible AI, she added, “requires participation upstream, not just evaluation downstream.”
Rafael Grohmann, assistant professor in U of T’s Department of Arts, Culture and Media, then turned attention to workers in the cultural industries, exploring how unions, cooperatives, grassroots collectives, and social movements are shaping how AI is used, negotiated, resisted, or refused in creative work.
Grohmann shared findings from an ongoing tracker of worker mobilizations around AI in arts, culture, and media, mapping more than 140 organizations across more than 30 countries and eight sectors. The project highlights how screenwriters, voice actors, musicians, visual artists, journalists, game workers, and other cultural workers are organizing around issues such as consent, compensation, credit, and the boundaries of acceptable AI use.
In the discussion that followed, the speakers emphasized that trustworthy AI depends less on any single technical fix than on the systems of accountability built around it. The conversation returned to the need for stronger operational safeguards, meaningful participation from affected communities and workers, and governance models that can adapt as AI moves from controlled pilots into messy, high-stakes real-world settings.
From left to right: Panelists Rafael Grohman, Laura Rosella, and Shingai Manjengwa.
Expertise, power, and AI futures
The second session, “Governing Intelligence: Expertise, Power, and AI Futures,” shifted from responsible development to the broader systems that shape AI governance. Moderated by SRI Graduate Fellows Lunjun Zhang and Kaushar Mahetaji, the session asked who governs AI systems, what kinds of expertise matter, and how narratives from industry, media, and policy shape the futures being built.
Kicking off the session, Zhijing Jin, an assistant professor in U of T’s Department of Computer Science at the University of Toronto and Canada CIFAR AI Chair at the Vector Institute, presented on democracy defense in the era of large language models (LLMs). Drawing on recent work by her lab and collaborators, Jin argued that much public attention to generative AI risk focuses on inaccuracy, cybersecurity, and copyright, while less attention is paid to political stability, human rights, historical accuracy, and democratic values.
Her work proposes evaluating LLMs not only along a left–right political spectrum, but also along a democracy–authoritarianism axis. Jin discussed research examining how models respond to prompts about political role models, human rights, and historical narratives, as well as the broader social risks created by generative AI.
A key concern, Jin said, is the asymmetry between producing misleading information and correcting it. Discussing a fake robocall deployed in the United States before an election and generated to sound like Joe Biden, she noted that synthetic media can be created quickly and cheaply, while verification and correction can take far longer.
“This is an example of information that is cheap to produce but expensive to verify and correct,” Jin said. That asymmetry, she argued, can reshape how citizens consume information, participate in democratic systems, and maintain shared understanding.
The session also heard from Vanessa Richter, a postdoctoral researcher at the University of Bremen, who examined how AI futures are shaped through public narratives and sociotechnical imaginaries. Richter argued that AI is not a fixed object with a single inevitable trajectory. Instead, it is continually defined through public discourse, policy debates, corporate communication, and media narratives.
“Understanding AI means also understanding who defines it,” Richter said. The question of who gets to define AI, she added, is also a question of “who gets to shape collective visions and technological futures and their governance.”
Concluding the panel was Peter Lewis, Associate Professor at Ontario Tech University, who discussed trust, bias, accessibility, and the development of AI systems that can work well as part of society. A Canada Research Chair in Trustworthy Artificial Intelligence, Lewis’ work bridges foundational AI research with human factors such as norms, values, social action, and trust, and his talk emphasized challenges in maintaining advanced technical systems that can match the fluid and reciprocal nature of trust.
SRI Graduate Fellow Mai Ali discussed her research during the workshop poster session.
Together, the sessions underscored a central theme: responsible AI futures cannot be designed by any single field, institution, or technical method alone. They require operational safety cultures, community participation, worker power, democratic accountability, and careful attention to the narratives that make some futures seem inevitable while obscuring others.
For SRI’s graduate fellows, the workshop offered an opportunity not only to showcase research, but to convene a broader conversation about the forms of expertise and authority needed in an AI-shaped world. Across healthcare, cultural labour, democratic systems, and public imaginaries, speakers returned to a shared conclusion: AI governance must be built before, during, and after deployment, with meaningful participation from the people and communities whose futures are at stake.
Want to learn more?
Browse stories by tag:
- AI
- AI safety
- AI trust
- Computer science
- Computer security
- Copyright
- Cybersecurity
- Data
- Democracy
- Economics
- Education
- Engineering
- Ethics
- GPO-AI
- Governance
- Health
- Human Rights
- In the Media
- Jobs
- LLMs
- Law
- Normativity
- Philosophy
- Political Science
- Privacy
- Privacy Series
- Psychology
- Public Policy
- Recommenders
- Regulation
- Religion
- Reports
- Trust
- Workshops
