Translating Relational Behaviours into Digital Processes

This image, and below, previous research at Royal Holloway has translated insight on relational services into the design of digital interactions.

PPA research has shown that reliance on digitised automated decision-making (ADM) means the loss of the social and behavioural structures that surround and support the decision-making process more generally. 

ADM is not able to ‘help’ or ‘explain’ a decision to an operator. In these terms, the potential explainability of an algorithm is not a characteristic of the algorithm, but can only be found through an understanding of the relationship it has to the surrounding and supporting social structures. 

The question of how to design ADM should be seen as an analogue problem rather than a purely digital one. Relationships and social constructs are therefore the key to the design problem. The ADM design process should not stop at seeking to make ‘better’ digital artefacts, but should generate new ways to represent the fact that relationships and social constructs carry with them different ways of thinking.

What is thrown away when you turn decision-making into a purely digital process? Translating human behaviours into digital processes is never straight-forward, since social relationships, behaviours and constructs are not rule-based, as algorithms are. In very many circumstances there is a need to understand where behaviours might have been mis-translated into digital form, breaking the sense of a system, and potentially becoming either system-breaking rather than system-supporting. This is why human agency is needed to help make sense of the interface between two systems – one being analogue and the other being the relationship between the digital, humans and algorithms.

Without this agency, there is always the danger that a digital conversational agent, for example, might make inadvertent and sensitive disclosures, and create other unforeseen potentially adverse effects on those they are purporting to help.

ADM requires a bottom-up approach that allows designs to be emergent from the necessarily varied problem space. Algorithm design can be seen as a subset of a much bigger problem: how to design technology for use in highly varied social problem spaces, among diverse social relationships, among a plurality of behaviours and social constructs. From the outset, ADM design cannot be conceived of as a top-down or pre-planned response to the richness and variety of social practices.