A New Paradigm for Algorithmic Decision Making and Policy Design

Design principles for algorithmic decision-making and policy design should include a transparency about what the aims of the system are. Here we visualise how the design of such systems might include a feedback loop between policy makers, technologists, refugee groups and refugees. Diagram courtesy of Parallel Systems.

Policy design can be described as the modelling of the society that we would want to see, or the ‘future subjunctive’ expression of it – defined by the Oxford English Dictionary as statements ‘used to express a wish, command, exhortation, or a contingent, hypothetical or prospective event’ (Oxford English Dictionary). Policy design can be informed by ‘backcasting’ from a desirable future state, working backwards from there to determine which policy measures would be required to reach that desired end-state.

In the case of energy policy, for example, there is the suggestion that backcasting can lead to the description of ‘soft’ energy policy paths – ones that carry the richness of a desirable end-state description, one that can be attained through softer pathways than might otherwise be used, and that become apparent because of these richer pictures of futures we would want. 

PPA develops upon this insight, and asks what the mechanisms would be for discovering new pathways towards policy goals. How can the complexities of problem spaces be explored in such a way that new areas of opportunity are discovered through interacting with data algorithmically? The argument being explored by PPA and our partners is that any decision-making process that has an impact on everyday safety, security and wellbeing should reflect and to some extent replay the dynamic and layered nature of the social problem space. The properties of these spaces are not fixed, but are relational, and creatively designed ADM systems should be able to take advantage of recent technological advances in simulation and hence be able to process data and provide decisions without losing sight of this relationality. Working with these dynamically shifting problem spaces means having thinking tools that are capable of presenting snapshots of subjunctive futures – a form of augmented data analytics which will have a bearing on policy. The goal set by PPA is that data and decisions should be seen and evaluated through the lens of the everyday stories emerging from this data. The type of algorithmic interaction envisaged by PPA is what will lead to new ‘soft’ resettlement policy paths.

How can this goal be achieved and what would it be like to work with such an ADM system? Put this question another way, how would a gamer want to approach this problem? One gamer reflected on the visual and action mechanisms of the game as a way to put oneself in a different position and to see the problem in a different way: ‘It's almost a visualisation tool for understanding, seeing and experiencing. Perhaps that's better, an experience tool' (reported by James Ash, writing on the game Crysis). Worldbuilding and non-static interaction with that world, especially if viewed through the lens of familiar everyday concerns, is precisely what is absent from standard models of ADM. 

The PPA research initiative Voices of Tomorrow aims to provide a testing ground for possible worlds created by policy decisions, and possible narratives within those worlds, even if they are only minimally different from our own world. As science fiction writer Philip K. Dick reflected, the act of creating credible and robust possible worlds has the effect of increasing awareness of the things which do not persist beyond the realm of a story and those that do: ‘Reality is that which doesn’t go away when you stop believing in it,’ he wrote.

What if you could use data to more fully understand what the future holds for people resettled through semi-automated systems? Through joint work between designers parallel systems, Territory Studios and researchers at StoryFutures and the Information Security Group at Royal Holloway, we are exploring how the simulation of life-like synthetic humans might allow users of future decision-making systems to hear first-hand accounts from the future, informing decision makers by bringing the focus onto what matters most – the future life outcomes of real people in different resettlement settings.

Voices of Tomorrow helps the ADM system users to understand and appreciate the qualitative aspects of what life might actually be like in a variety of resettlement scenarios. Utilising new game engine tools (Unreal/EPIC) and AI technologies (GPT-3), the Voices of Tomorrow collaboration aims to provide a unique form of insight into complex questions about future scenarios for refugees. By considering a broader range of indicators, from both internal and external sources, organisations can make more accurate predictions of outcomes, and help decision-makers and policy-makers reflect on the desirability of alternative outcomes and policy goals. As policy objectives change, with interactive next-generation thinking tools like the Voices of Tomorrow prototype ideas it becomes possible for policy-makers and decision-makers to continuously calibrate and renew a picture of what a successful application of policy might look and sound like. 

Voices of Tomorrow takes up the challenge of redefining how decision-making systems are to be optimised, in areas such as quality of life, economic benefit or the environmental impact of different alternative futures for refugees and evacuees. Voices of Tomorrow aims to help all stakeholders grasp the human impact of decisions that are being made.  

For more on the background research that has led to this approach, see: www.refractions.org.uk

Collaborators

Parallel Systems: 
https://parallel.systems/

Territory Studios: 
https://territorystudio.com

StoryFutures:  
https://www.storyfutures.com

Backcasting and Transition Design. Based on Irwin, Tonkinwise and Kossoff, 2016.