Session purpose
This session explored whether AI can help think tanks strengthen trust, rather than simply increase productivity. It asked how think tanks can use AI in ways that preserve credibility, transparency, ethical standards and human judgement — the very qualities that make policy research organisations valuable.
The session combined conceptual discussion with practical design work. It examined responsible AI standards, AI-enhanced search, AI agents, organisational legitimacy, disclosure, human oversight, and the governance features needed to ensure that AI tools support trust rather than undermine it.
Main conveners and contributors
This session was convened by:
- Sonia Jalfin, Sociopúblico
- Noel Gruber, CIGI
The session also included contributions from John Schwartz, who focused on AI-enhanced search and vector embeddings, and from participants who worked in groups to design AI agents for think tank workflows.
Main presentation and framing
The session began with the question of organisational legitimacy. Noel introduced three dimensions of legitimacy that should guide AI use in think tanks:
- Pragmatic legitimacy: Does AI provide real benefit to users?
- Moral legitimacy: Is AI being used fairly, responsibly and with attention to bias?
- Cognitive legitimacy: Does AI use feel appropriate for a think tank and its role?
This framing positioned AI adoption as more than a technical choice. It is also a reputational and institutional decision. Poorly designed AI tools, the hidden use of AI, or unreliable outputs can quickly damage trust among staff, funders, policymakers, and wider audiences.
Noel also introduced a forthcoming trust standard for responsible AI in policy research, based on three pillars: human leadership, transparency about AI’s role, and responsible analysis using credible sources.
Main debates
1. AI must support legitimacy
A central debate was whether AI adoption strengthens or weakens think tank credibility. The answer depends on use and governance.
AI can support legitimacy by helping users find relevant knowledge, improving workflows, making processes more transparent, and supporting better judgment. But it can damage legitimacy when it produces misinformation, hides sources, introduces bias or gives the impression that the organisation is outsourcing its intellectual authority to a machine.
2. Transparency is non-negotiable
Participants stressed that think tanks need to be clear about when and how AI is used. Disclosure is not only an ethical issue; it is central to maintaining trust.
This applies to research outputs, search tools, automated briefings, AI agents and internal knowledge systems. Users need to know what AI has done, what humans have reviewed and who remains accountable for the final product.
3. Human leadership is stronger than “human in the loop”
The session challenged the common idea that AI is acceptable as long as there is a “human in the loop”. The proposed standard instead emphasised human leadership.
This means researchers and policy experts should define the problem, select sources, set boundaries, interpret the outputs, and remain responsible for quality. A human should not merely approve an AI-generated answer at the end. In think tank work, judgment must lead the process.
4. AI-enhanced search is an immediate opportunity
John Schwartz argued that many think tanks still have poor website search, making valuable archives difficult to use. AI-enhanced search using vector embeddings can help users search by meaning rather than by exact keywords.
This is relatively low-cost and can make existing research much more accessible. It can also provide the foundation for personalised briefings, internal and external content summaries, media alerts and direct AI querying of an organisation’s research database.
5. AI agents can support institutional work
Sonia introduced AI agents as tools that can perform defined multi-step tasks, rather than simply respond to individual prompts. Participants explored several possible agents for think tanks, including:
- a conference travel assistant;
- a lessons learned interview agent;
- a collaboration pattern finder;
- an internal research synthesis agent;
- media monitoring agents;
- legislative tracking agents.
These examples showed how AI could support knowledge management, policy monitoring, organisational learning, event preparation and collaboration. But they also raised questions about reliability, auditing, data sources and human oversight.
6. The corpus determines the quality of the output
The discussion repeatedly returned to the importance of source quality. AI tools are only as trustworthy as the data and documents they draw from.
For think tanks, this means carefully defining the corpus: which publications, databases, media sources, legislative records, internal documents and expert profiles the AI can use. Poor inputs will produce weak outputs, even if the tool itself is sophisticated.
7. Trust must be designed into the tool
The workshop introduced a trust sticker framework to help participants design AI agents with trust principles built in from the start. These included transparency, security, identity and voice, and governance.
Participants prioritised features such as auditable processes, identifiable human owners, controlled updates, user-specific customisation and clear communication about AI involvement. The message was that trust cannot be added at the end; it must be part of the design process.
8. AI adoption is organisational change
The session also recognised that using AI well requires a change in organisational habits. Staff need to make their assumptions, criteria and reasoning more explicit. This can be uncomfortable, but it is necessary if AI is to act as a useful co-worker rather than a black box.
AI adoption is therefore not only about tools. It is about culture, workflows, training, accountability and shared standards.
Main takeaways
- AI can become part of trust infrastructure only if it is transparent, human-led, ethical and governed.
- Think tanks should not use AI simply to produce more content faster. They should use it to improve access, relevance, learning, accountability and engagement.
- Responsible AI use depends on credible sources, clear disclosure and visible human accountability.
- AI-enhanced search is one of the most immediate opportunities for think tanks because it makes existing knowledge easier to find and use.
- AI agents can support media monitoring, legislative tracking, conference preparation, internal synthesis and lessons learning, but they need quality control and governance.
- The quality of AI outputs depends heavily on the quality and boundaries of the data corpus.
- Trustworthy AI design requires auditability, source transparency, human ownership, security and clear communication to users.
- The shift from “human in the loop” to “human-led” is important. Think tanks must retain responsibility for framing, judgement and final outputs.
- The strongest message of the session was that AI should help think tanks become more trustworthy, not just more efficient.
Relevant resources
- Guidance for Responsible AI standards in policy research and public services:
- Resources on Responsible Use and AI disclosure policies:
- Resources on AI governance, transparency and auditability:
- OTT resources on AI, trust and think tanks: