If there's one fixture nearly every public dialogue shares, it's the expert. Scientists, academics and specialists are brought in to inform participants, and there's usually a Q&A where the public can ask questions directly. It's treated as essential - and intuitively it should be. 

We think commissioners of deliberation should be able to make decisions grounded in evidence. So we’ve tested whether hearing from an expert and getting to interrogate them changes things. Does it have an impact on what people conclude, how much they enjoy the session, and how worthwhile they judge the whole exercise?

The somewhat surprising result: the Q&A made almost no difference.

In another experiment, during our AI Social Readiness project, we found that well-run online deliberation held its own against in-person. This time we turned to the expert.

So we tested it. Across the same programme of 36 workshops on AI in public services - 280 participants, broadly reflective of the UK, randomly assigned - half the sessions included a Q&A with an independent AI expert and half did not, with the same materials, facilitators and platform throughout. For participants, the evidence was delivered through nine short animated videos, built with experts and independently checked for balance and accuracy.

We looked at the same four outcomes as before - deliberation quality, enjoyment, how effective participants judged the process, and how far their attitudes to AI shifted.

On three of them, we found no meaningful difference between workshops with a Q&A and those without. Deliberation quality held at 4.76 out of 5; perceived effectiveness of the process; attitude shift at a steady +0.39 points on a 1-5 scale across five attitudinal statements, which included trust, knowledge and comfort with public sector AI. (As in our first experiment, these scores sit near the top of their scales, so there was limited room for a gap to open up.) The only movement was in enjoyment, where the Q&A was associated with a slight increase, but it was small and not statistically significant (+0.20 points on a 1-10 scale, p=.27), so we wouldn't read much into it.

The first reading is the cautious one: an effect could exist that our design was too small to detect. But there's possibly a more interesting explanation, and it's specific to how we designed the AI Social Readiness process. We had conducted a lot of user research to anticipate the questions participants would want to ask - and many were already answered by the videos they'd watched. The expert, in a sense, had already been in the room.

That's the takeaway we 'would offer commissioners: engagement with expert evidence doesn't have to arrive through a live presentation and Q&A. Done well, other formats can do the same job. For anyone who 'has spent weeks lining up speakers, rewriting jargon-heavy slides, or scrambling when a key voice pulls out the day before, that may be welcome news.

Two honest limits, though. Our four measures mostly capture how participants experienced the deliberation - its quality, their enjoyment, the value they saw in it. Although it didn’t find a difference in the extent to which their attitudes changed, this experiment didn’t measure whether participants ended up better informed or reasoned more sharply. And this was one expert, on one topic. Perhaps on a different topic, or where there were contested expert views, the live Q&A might earn its place more clearly.

These findings do not mean experts don't belong in deliberation. There are good reasons to involve them beyond informing participants - not least so they hear, first-hand, the quality of public thinking on their subject. Experts are often surprised by it, and may feel a stronger sense of responsibility to act on what they hear when they've been part of the conversation. (That's an experiment we'd love to run next.)

If you're commissioning public deliberation and want to talk about designing it for both quality and value - online, in person, or a blend of the two - we'd love to hear from you. Please email us at [email protected].

Part of a series of experiments from Nesta's Centre for Collective Intelligence.