The Future of Policymaking: Can AI Replace the Human Touch?

On a bright June morning in 2026, over 70 policymakers, researchers, and knowledge brokers gathered virtually for a thought-provoking discussion: How is generative AI changing the way we create evidence-based policy? The webinar, hosted by the Community of Practice on Evidence-Informed Policymaking (EIPM), brought together Nora Revai from the OECD and Julien Gossé from the European Commission’s Joint Research Centre (JRC) to explore AI’s growing role—and its limits—in the world of policymaking.

The Promise and Pitfalls of AI in Policy Work

Nora Revai opened with a provocative scenario: "A minister walks into your office and says, ‘Teachers are quitting—we need solutions now!’ An hour later, your AI assistant delivers a fully fleshed-out policy brief, complete with evidence." Sounds ideal, right? The question is: Can we trust it?

AI is already transforming how knowledge brokers, those who bridge the gap between research and policy, do their jobs. It can screen thousands of research papers in minutes, extract key data, and even draft policy recommendations. Yet, as Nora pointed out, AI still struggles with distinguishing different meanings of the same term in different literatures (or different terms for the same meaning), critical appraisal, and contextual understanding. It might confidently cite a study, only for a human to realise it’s misrepresented or outdated. And while AI can synthesise evidence, it can’t yet build trust the way a human can.

Where AI Shines—and Where It Falls Short

Julien Gossé took the discussion further, asking: When should we use AI, and when should we rely on human judgment? In his view, it’s about collaboration, not replacement.

A 2024 study he cited found that humans + AI working together outperformed either alone. In a bird classification task, humans achieved 81% accuracy, AI hit 73%, but together, they reached 90%. The lesson? AI amplifies human strengths when used wisely.

But not all tasks are equal. Julien broke down the policy brief-writing process, showing where AI helps (and where it doesn’t):

• Defining the problem? AI can map trends, but humans must align stakeholders.

• Searching for evidence? AI tops at speed, but humans judge relevance.

• Drafting recommendations? AI can structure ideas, but humans ensure they’re feasible, ethical, and context-aware.

The Big Question: What Should AI Never Do?

The most debated question of the day: What tasks should AI avoid, even if it could do them?

Nora and Julien agreed: Trust and relationships are non-negotiable. AI can’t (yet) replace the human ability to negotiate, persuade, or understand unspoken political realities. As one participant put it, "Policymakers expect critical thinking; something AI can’t fully replicate."

But the discussion didn’t end there. Some argued that AI could eventually analyse norms and values (if fed the right data). Others warned of over-dependence, citing risks like "automation bias" (trusting AI too much) and "skills erosion" (losing human expertise).

Six Rules for Working with AI Responsibly

Julien left the audience with six guiding principles for human-AI collaboration:

1. Understand AI’s limits—don’t assume it’s always the best tool.
2. Experiment first—test AI in small doses before scaling up.
3. Stay in control—AI should assist, not dictate.
4. Keep thinking critically—don’t let AI replace your judgment.
5. Consider the bigger picture—how does AI affect teamwork and long-term goals?
6. Share knowledge—the best insights come from collective learning.

What’s Next?

The webinar made one thing clear: AI is a tool, not a replacement. Knowledge brokers must adapt, experiment, and set boundaries to ensure AI serves policymaking..

As we look ahead, the community of practice on EIPM will keep exploring these questions. Our next webinar will dive into how AI is reshaping policy jobs. Stay tuned!