Dr Michael Hallsworth
Chief Behavioural Scientist
This is the fourth and final piece in our series explaining the four big shifts in behavioural science that BIT is aiming to achieve.
In this series, I’ve explored how behavioural science must evolve – by thinking in systems, understanding social drivers of behaviour, and adopting adaptive approaches. The final shift in behavioural science looks at who acts:
“Be more participatory. Involve people in behavioural science work, support them in shaping projects, and help them apply behavioural science themselves.”
Over the past 15 years, BIT and others have emphasized the importance of exploring the context from the perspective of those using and delivering services – talking to them, seeing what they saw and applying those insights to unlock an issue or rethink it entirely. One of the most famous nudges, shifting UK pensions from opt-in to opt-out, emerged from an extensive programme of public engagements, including a National Pension Day involving 1,000 people.
In this piece I show how we can push further on four aspects of participation: consultation, deliberation, co-design and co-creation. You’ll see how participation can be a central way of rethinking behavioral science – particularly because of recent advances in artificial intelligence.
1. Consultation
The simplest way to increase participation is to ask people for their thoughts. Recently, that principle has led researchers down an interesting path: gathering views on behavioral science interventions themselves. We’ve learned that many nudge-type policies seem to be more popular than sceptics assume, and that most people simply seem to judge a nudge based on whether they agree with the policy’s goal, rather than the use of a nudge itself. In a way, this should not be surprising: most people do not disagree with the use of taxes and legislation as such; they focus on what they try to achieve.
But we’re still way off being able to predict what people will find acceptable. For example, one study asked people to rate the acceptability of nudge-type policies. Half the descriptions mentioned that the people affected by the policy had been consulted on it beforehand. Surprisingly, it seems that the policy was seen as less acceptable if consultation had taken place. Does that result hold true in the real world? How often? Similarly, there are few public studies on how much people want interventions to be tailored – and using what data. These are just a couple of the crucial questions that targeted consultations could answer.
The studies I just mentioned proactively ask people for responses, but another strategy may be to look where people are already offering up their thoughts. For example, millions of people take the time to write complaints. They can act as crucial leading indicators of problems – but only if they are analyzed. BIT worked with the London School of Economics to develop new ways of studying complaints made to the UK’s National Health Service, in order to reveal problematic patterns of behaviors. Michael Luca from Harvard has shown that analysing the billions of consumer reviews on platforms like Yelp can offer similar insights. There’s a massive new opportunity to combine behavioral science with the power of large language models here.
2. Deliberation
Deliberative forums go beyond asking what people think. They create spaces where people can develop their thinking through discussing the details and trade-offs that are core to an issue. In fact, if behavioral scientists design forums with a sustained, in-depth focus on evidence and rules that respect diverging opinions, they can change behaviour.
When BIT ran the largest ever Community Forum (6,300 people, 32 countries and 23 languages), participants often expressed surprise at how civil and substantial the interactions had been. They seemed to reset expectations about what an online discussion could be. BIT found a similar effect when it paired up people with strongly opposing views to have a Zoom chat with each other. AI offers the chance to improve things further – at least one study shows that AI mediators were more effective at reducing division in deliberating groups than human ones.
The benefits of deliberation are biggest when there’s a genuinely open or knotty question, with contested values and nuances to the trade-offs. That was true of the first forum we ran (on obesity policy back in 2015) and it’s true of the one we published a couple of weeks ago (on what counts as “fair” in gambling). The task of negotiating between radically different perspectives also helps behavioral science move away from a single, narrow concept of “rationality”.
Given these benefits, behavioral scientists may want to focus on finding new ways of helping people to engage in deliberation. We know that enthusiasm for different kinds of participation (e.g. participatory budgeting, citizens assembly) varies by group. We also know that, when asked, people state that lack of perceived opportunity is the biggest barrier to greater engagement. But will that play out in practice?
3. Co-design
Co-design invites people to get involved in designing policies or products, rather than just reflecting or advising on them. Creation comes from wider exchanges, which try to create more effective, tailored, and appropriate interventions that respond to a broader range of evidence.
In one study, researchers were looking to increase the use of a bank’s subscription management tool. They got experts to design a message to customers but also crowdsourced message ideas from customers. The four crowdsourced messages were more effective than the expert-generated message at getting people to open the email offer. 63,500 more people opened the message because the bank decided to expand the source of ideas.
How powerful could these crowdsourced messages be? Behavioral scientists don’t gather them often enough to know. When they do, there are glimpses that a fusion of experts and peers could be optimal. One study found that expert messages worked better at increasing physical activity for those beginning to exercise, while peer-designed messages were more effective for those who were already doing some exercise. Does this hold true in other contexts? We just don’t know yet – but we need to find out.
As we say in our manifesto, current debates in this area are a bit simplistic. Co-design may be a game-changer in some situations that are marked by social or symbolic distance between policymakers and communities, researchers and users. Or it may just form another burden put on people being researched. If behavioral science is a lens, that means those who are being researched can and should look back through it. That idea leads onto my final point, which allows us to zoom out, rethink some core principles, and chart new paths.
4. Co-creation
Let’s start by distinguishing between co-design of a defined “intervention” (policy/product/service) and a more open-ended process of co-creation. In co-creation, people may be actively producing the policy or product as they engage with it. That can make participation itself “the intervention”. For example, researchers have found that engaging with an AI chatbot can reduce conspiratorial beliefs.
This idea is not new. Deliberative forums have shown similar effects, and many psychological therapies are built on the same principle. But creating these interactions in one-on-one or small group settings has traditionally been labor-intensive. AI changes the game, allowing structured yet responsive interactions to happen at massive scale.
We need to understand the true significance of this shift. Co-creation won’t just be limited to actively producing a preset “intervention” – nor should it be. Interactions can mean that something new emerges that was not anticipated by any designer. For instance, a local authority’s AI assistant, created to help residents understand new recycling rules, could end up as a source of crowdsourced ways to improve that process itself – and seeing the local authority respond to feedback sparks more recycling by residents.
Of course, these experiments are already happening: millions of people have embarked on personal journeys interacting with AI. So where does this mass co-creation leave behavioral science? The answer pulls together several strands that have been threaded through this series. For a start, even when people are creating interventions “from scratch”, they are still interacting with AI systems that are based on human designs. Behavioural scientists can add a lot here by providing expert input on how humans interact with the “choice architecture” of artificial intelligence; BIT is actively working on that question.
Second, AI will accelerate the process of broadening behavioral science beyond a crude model of experts creating an intervention that is applied to others. The stability and control implied by that model sits badly with a world of mass co-creation. So behavioural scientists should also be trying to shape the ‘rules of the game’ (first big shift), understand the social and cultural factors that drive the interactions (second big shift), and adapt to the unexpected outcomes that they produce (third big shift). If behavioural scientists can see themselves more as facilitators of change created by others, they will unlock a more effective and sustainable future for the field.
