Participatory AI futures: lessons from research in climate change

by Helena Hollis and Dr Jess Whittlestone

What do we, as a society, want from artificial intelligence? This is an enormous question, one we think isn’t being asked enough. It certainly isn’t being asked of enough people. If we want AI to benefit society as a whole, then we need to draw on the perspectives of a wide range of groups when envisioning the future of AI, and in making decisions about how we get there.

This is the aim of participatory futures methods: to mobilise far larger numbers of people in thinking about the future, to ensure it isn’t driven by a narrow set of interests (Smith and Peach). While the notion of participatory futures in AI is relatively new, there is an extensive body of research in climate change which includes future-oriented participatory approaches. In this post, we discuss what the field of AI can learn from how participatory futures have been used in the climate domain, drawing on the findings of several review papers (listed at the end of this post). Where we make general claims without specifying sources, we are drawing from commonalities across these review papers. Where we reference sources other than these reviews, we have provided hyperlinks. This is an initial, informal review of the literature, but we hope we can point to some useful directions and considerations.

Comparing participatory futures in AI and climate change research

How relevant is participatory futures research in the climate domain for the kind of research we might want to do with AI?

There are several ways in which thinking about the future possibilities of AI development may be analogous to thinking about climate futures:

• Both AI and climate change are likely to have global, transformative impacts on all parts of society, meaning all parts of society should have a voice in decision-making.
• In both cases, there is need for action on governmental, business, and public levels to mitigate harms, and there may be challenging tradeoffs that bring different perspectives and interests into tension.
• In both cases, expert understanding is essential for many aspects of prediction and decision-making, but local communities are also likely to have expertise when it comes to understanding actual and potential impacts.
• There is also reason to think that the worst impacts of both climate change and AI will (at least initially) disproportionately affect groups that are already poor and vulnerable, making listening to those groups especially important.

However, there are also some disanalogies between the two domains which we should bear in mind when drawing lessons:

• The vast majority of reviews we looked at addressed climate adaptation — i.e. how communities can preserve their livelihoods and ways of life in the face of climate change — rather than mitigation i.e. preventing negative climate impacts. This reflects where we are in the climate emergency (the latest IPCC report makes stark the damage that has already been done). By contrast, we are not yet at the point of adapting to the impacts of AI, and understanding and mitigating harms should be more of a focus. This suggests that some specific aspects of adaptation-focused methods may be less relevant for participatory work on AI impacts.
• Understanding and predicting the impacts of climate change is a more mature field than understanding and predicting the impacts of AI; this may suggest that the relative importance of expert vs. community insights may need to be thought about differently.
• Participatory approaches in climate adaptation are often focused on rural, and frequently indigenous, people facing specific physical changes in their environment as the climate changes and destabilises. Different groups may need to be prioritised in terms of AI impacts.
• Since the impacts of AI are likely to be less focused on the physical environment, many of the methods used in climate change research are unlikely to translate well to AI studies (e.g. ‘transect walks’ in which local people take the researcher on a walk around their environment and point out key features, or calendar methods that track local weather events).

However, there may still be scope for creative re-thinking of approaches from the climate domain, even if they do not translate directly. For example, asking local people to walk researchers through key parts of their day, rather than through a physical environment, could help identify ways AI technologies may develop to impact everyday life. In addition, physical, localised changes and regional differences ought not be ignored altogether as AI develops. For example, walks around areas transformed by warehousing could tell us something about the trajectory of digital technology driven business change, which does have physical footprints. Not all methods will be transferable, but approaching them with a creative eye is always advisable.

Insights from the use of participatory methods in climate change research

The aims of participation

Reviews of participatory futures approaches in climate change research point to two broad aims of undertaking such research:

1. Improving knowledge and understanding of issues within communities
2. Improving decision-making by drawing on wider information and perspectives

This highlights two different ways of thinking about the goal of participatory research, which are not necessarily mutually exclusive, but will suggest ways of designing and evaluating studies. The first is to see certain forms of participation as an inherent good and necessary in democratic societies. This perspective mostly focuses on the benefits to participants themselves in terms of improved understanding and engagement with issues, as well as potentially giving them access to greater resources and networks with which to articulate their views. The second perspective focuses more on the role of participation in improving societal-level decision-making and outcomes, both by highlighting useful information and perspectives, and by adding legitimacy to decisions which may increase acceptance and compliance. Of course, these two aims are related, since improving participants’ knowledge and resources may often be an important route to them contributing to better decision-making more broadly. As we’ll discuss later, evaluating these different outcomes (and the relationship between them) can be very challenging in practice, and this is an area where research could greatly improve in future.

Methods used

From the papers we reviewed, future-oriented participatory approaches in the climate domain fall into two (overlapping) clusters:

• Scenario methods, which typically aim to generate multiple different possibilities for different futures that might arise. This can be done through visioning (projecting forward from now into a possible future), or backcasting (imagining a future possibility and then tracing a trajectory backwards to the present). Generating scenarios can be used as an exploratory method, to chart a range of possibilities, or as a more normative method aiming to evaluate specific policy outcomes. Scenario research can involve participation to very different degrees, with scenarios either made with participants, or handed to participants, with different levels of expert involvement.

For example: Beach and Clark (2015) ran three workshops with Yukon wildlife managers: first identifying drivers of change; second developing scenario narratives; third identifying responses to the different scenarios. Some lessons drawn from the process included that scenario planning helped the participants identify overlooked needs, and the use of scenarios enabled the application of traditional knowledge to planning.

• Risk assessments, which tend to take the present as a starting point and consider upcoming risks and how they might be addressed today. These methods are typically focused on risks faced by specific localised communities, and so often involve surveying the current state of affairs with a focus on livelihoods, then considering the adaptive capacity of a group in the face of different future possibilities. Community workshops in different forms are perhaps the most common approach.

For example: Aalst et al. (2008) describe a community risk assessment carried out by the Costa Rica Red Cross in a neighbourhood prone to landslides. Using approaches such as community maps, transect walks, and workshops and interviews, they found landslides were a concern for the community, but also identified other risks such as fire hazards. They were able to design solutions appropriate for the risks the community members had highlighted, and the community also designed their own initiatives.

Risk assessments can involve scenarios, and scenarios can be used to assess risk; these groupings merely help differentiate different potential aims.

There is also a wider family of research methods that seek public engagement and participation as their key purpose. These approaches are often framed as “knowledge generating”, and/or “co-production”, and typically involve diverse means of engaging participants such as workshops, panels, juries etc. These can also involve arts approaches that aim to explore participants’ decision making in light of different possibilities, such as serious games and participatory theatre.

Some of the approaches we found especially interesting in this space are dialogic, with phases shifting between expert and public input. For example, Cone et al 2013 moved between input from scientists, and the community, to model regional climate influences, culminating in a shared strategy for communication. The dialogic approach is appealing as it offers knowledge sharing in both directions between local and scientific communities.

Successes and benefits

As already discussed, the benefits of such research are often argued to lie in the experiences of the participants themselves, for whom participation could offer avenues of learning and empowerment. On one level, this is thought to promote individual development and behaviour change. For instance, in the Yukon example the wildlife managers described how the scenario discussion process helped them to visualise the different components impacting the wildlife in their region, giving them a better understanding of threats.

However, there are also benefits reported at a wider scale, with communities able to access greater resourcing and increase their networking with the governmental and business stakeholders brought into communication through the research. This is shown in the Costa Rica example, where the community both initiated their own responses to the risks they identified, but also received more targeted infrastructure development.

Further benefits are reported for the quality of the research output itself. Researchers can gain unexpected insights through local-level information they could only have accessed by engaging with the community. In the Costa Rica case, this is shown in the risks beyond landslides (which had originally motivated the research) that the community members identified. In the Yukon case, the researchers noted how First Nations participants had a different way of classifying scenario information which informed the development of different axes for mapping drivers of change. Such novel input from participants leads many reviewers to note the importance of such research where the future is highly unpredictable, as diverse voices can offer a broader range of possibilities where hard prediction is unobtainable. Through increasing the scope of future possibilities under consideration, surprises can be preempted.

Participants from local communities can also provide more insight into existing adaptation approaches already in use. Their existing experiences of what works and does not can be deployed in planning future approaches, which can help avoid maladaptive interventions being imposed. As McNamara & Buggy (2017) highlight, examples of maladaptive “command and control” approaches abound in cases where communities were not involved.

Limitations and challenges

Reviews note that the time and resources needed are often the biggest factors preventing more participatory research. The costly nature of this research also means there is a persistent lack of rigorous evaluation; as Aalst et al (2008) note, reporting by practitioners forms the backbone of the evidence base. Evaluating the impacts on not just participant understanding but also increased action and better decision-making is particularly challenging because it requires longitudinal follow up which tends to be costly. As a result, there is an acknowledged lack of clear evidence to support the causal chain from participatory research to increased action.

Beyond evaluation challenges, there is reason to expect a gap between participation and increased action. Hugel & Davies (2020) point out that there has been a long history of public engagement and activism with climate issues, but a gap between this and policy decision making, and action taken to curb global heating. Many studies comment on the divide between the local impact of participatory studies and the wider global context, as noted by O’Neill et al (2013) who point out the challenges in predictions where there are complex linkages across scales.

There are also deep issues to be considered in the intrinsically hierarchical structuring of research, in which the researcher engages with participants from a position that unavoidably carries a power imbalance. When involving different stakeholders with different levels of power (such as governments and businesses, alongside local people), these power imbalances become even more concerning, and there is no simple recipe to follow in order to resolve them.

Stemming from these issues, outcomes of participatory research can be negative. Reviewers of the research (e.g. Hugel & Davies, 2020) warn that projects reaching out to disenfranchised communities can be used to quell their discontent, providing a mere semblance of listening, offering a false promise of action, if those in power do not ultimately deliver real change as a result of the community input. While legitimising decision making by increasing democratic input can be a major benefit of participatory research, it can also be used to rubber-stamp decisions by putting on a display of democratic input-gathering. The ethics of these research approaches are therefore highly complex.

Finally, the prediction-making facets of these studies also have the potential for doing harm, and reviews of these studies warn of creating a false sense of predictability (e.g. Sutherland & Woodroof, 2009). Preempting surprises is a worthwhile goal, but not all surprises can be preempted, and it is therefore essential to communicate this to everyone involved.

Lessons for using participatory futures in AI research

Despite the challenges, we believe that participatory methods are needed in the AI futures domain. An argument worth keeping in mind is as AI develops and policy pertaining to its development becomes ever more pressing, some groups — those with the highest levels of wealth and influence — are participating in the processes that yield policy decisions already; Sarzynski (2015) points out how this has been the case in climate policy lobbying. Without an active effort to engage with communities that are not already so well connected, the voices of the most privileged will thus disproportionately shape the decisions being made. This should not only concern us in principle, but also in terms of the practical implications of letting vested interests shape policy, as we’ve seen in the influence of the fossil fuel industry on climate policy making.

We suggest that participatory futures research in AI should aim to:

• Conduct more rigorous evaluation against clear aims. A clear limitation of many of the reviews we looked at is the lack of rigorous evaluation, especially when it comes to assessing whether participatory research leads to improved decision-making or increased action. If we want to take participation seriously as a method for understanding and mitigating harms from AI, we must start with a set of clear aims and methodologies for evaluating the extent to which those aims have been met.
• Consider the relevance of different perspectives and expertise for different goals. A theme that came up implicitly in many of the papers we reviewed is how to balance the importance of different perspectives, and particularly how to balance the need for subject-matter expertise with the importance of diverse perspectives. This balance will depend on the goal of the study: if the goal is to understand a wide range of possible impacts of AI, for example, a dialogic approach where domain expertise and community perspectives can build on one another may be particularly effective; whereas if the goal is to understand the impact of a new technology on a particular demographic, it will be more important to really understand the experiences of the relevant communities.
• Think carefully about how to manage power imbalances. A concern raised in the climate literature is that an obvious sense of ‘hierarchy’ between participants and a researcher may make it difficult to obtain the most useful insights, and can leave communities feeling disenfranchised and even exploited. In the context of AI, we want to make sure participatory research can empower rather than doing the opposite. Seeking expert advice on how to avoid this and ensure participants feel comfortable and valued, as well as clearly communicating the aims of studies, will be important.

This review of reviews gives some broad ideas that could be taken from the climate domain and adapted in AI futures research. However, there is far more to be gained by engaging in depth with participatory approaches in this and other fields. Just as AI impacts intersect with many other global challenges, so work on AI should draw from many other spheres.

The authors:

Helena Hollis is a PhD researcher in the UCL Department of Information Studies, and she’s also coordinating a shared UCL and British Academy project on AI and the Future of Work.

Dr Jess Whittlestone is Senior Research Fellow at the Centre for the Study of Existential Risk and the Leverhulme Centre for the Future of Intelligence, both at the University of Cambridge. She is working on the long-term societal impacts of developments in artificial intelligence, and how those impacts can be made positive.

List of reviews of participatory research on climate change including a future focus:

Aalst et al. 2008, Community level adaptation to climate change: The potential role of participatory community risk assessment, Global Environmental Change, 18, pp.165–179, https://doi.org/10.1016/j.gloenvcha.2007.06.002

Adger et al. 2003, Adaptation to climate change in the developing world, Progress in Development Studies, 3(3) pp.179–195, https://doi.org/10.1191/1464993403ps060oa

Bengstong et al. 2012, Strengthening Environmental Foresight: Potential Contributions of Futures Research, Ecology and Society, 17(2), https://www.jstor.org/stable/26269050

Flood et al. 2018, Adaptive and interactive climate futures: systematic review of ‘serious games’ for engagement and decision-making, Environ. Res. Lett., 13, https://doi.org/10.1088/1748-9326/aac1c6

Flynn et al. 2018, Participatory scenario planning and climate change impacts, adaptation and vulnerability research in the Arctic, Environmental Science and Policy, 79, pp.45–53, https://doi.org/10.1016/j.envsci.2017.10.012

Gidley 2016, Understanding the Breadth of Futures Studies through a Dialogue with Climate Change, World Future Review, 8(1), pp.24–38, https://doi.org/10.1177/1946756715627369

Gidley et al. 2009, Participatory futures methods: towards adaptability and resilience in climate‐vulnerable communities, Environmental Policy and Governance, 19(6), pp.427–440, doi:10.1002/eet.524

Hugel & Davies 2020, Public participation, engagement, and climate change adaptation: A review of the research literature, WIREs Clim Change, 11(e645), https://doi.org/10.1002/wcc.645

McNamara & Buggy 2017, Community-based climate change adaptation: a review of academic literature, Local environment, 22(4), pp.443–460, https://doi.org/10.1080/13549839.2016.1216954

O’Neill et al. 2013, Methods and Global Environmental Governance, Annu. Rev. Environ. Resour. 38, pp.441–71, 10.1146/annurev-environ-072811–114530

Sarzynski 2015, Public participation, civic capacity, and climate change adaptation in cities, Urban Climate, 14, pp.52–67, https://doi.org/10.1016/j.uclim.2015.08.002

Sutherland & Woodroof 2009, The need for environmental horizon scanning, Tends in Ecology and Evolution Update Forum, 14(10), pp.523–527, https://doi.org/10.1016/j.tree.2009.04.008

vanVuuren et al. 2012, Scenarios in Global Environmental Assessments: Key characteristics and lessons for future use, Global Environmental Change, 22, pp.884–895, https://doi.org/10.1016/j.gloenvcha.2012.06.001

Vervoort & Gupta 2018, Anticipating climate futures in a 1.5 °C era: the link between foresight and governance, Current Opinion in Environmental Sustainability, 31, pp.104–111, https://doi.org/10.1016/j.cosust.2018.01.004.

Whitmarsh, et al. 2013, Public engagement with climate change: what do we know and where do we go from here?, International Journal of Media & Cultural Politics, 9(1), pp.7–25, https://doi.org/10.1386/macp.9.1.7_1