Workshop report: Diagramming AI in the Street
AI in the street is a collaborative project that explores the messy reality of AI as encountered in the street, in the form of automated vehicles, delivery drones and surveillance infrastructure. To make AI visible for everyday publics, researchers work together with artists, designers and local partners to create street-based interventions in 5 cities in the UK and Australia. The aim is to ground understandings of AI in lived experiences.
Here below we report on the first workshop of the project, in which project participants came together to create Diagrams of AI in the Street.
Introduction
For our very first AI in the Street workshop, which took place at the University of Warwick in March 2024, project researchers came together to review and model the presence of AI in the street.
Participants joined from the Universities of Warwick, Cambridge, Edinburgh and Monash and had varying backgrounds, from the arts to sociology, design research and computer science. Before the workshop, each participant was asked to bring with them an example of an AI technology that is currently in use in the street and to consider who and what is involved in the situations that the presence of their chosen AI in the street gives rise to.
Their examples provided the basis for a mapping of “AI in the street" during the workshop, which we present below.
Approach to Diagramming AI in the Street
To begin exploring AI in the Street during this initial workshop, we then adopted a case-based approach. We wished to avoid the abstraction that utilising “observation” of AI may result in.
In choosing their example of AI in the street, participants were encouraged to define AI broadly, in terms of the automation and automat-ability of human, social and institutional tasks. We asked them to consider the following: who and what is involved in the use of their chosen AI in the street? What conditions allow the humans and non-humans involved in their case to observe or sense the presence of AI and of one another.
In mapping AI in the street and the entities associated with it in this way, we draw on established methods in social studies of technology and society, such as the creation of socio-technical graphs and the mapping of “actor-networks” developed by Bruno Latour and colleagues. At the same time, it is clear to us that such classic approaches to mapping socio-technical networks have significant limitations when it comes to diagramming AI in the street. Actor-network mapping does not enable us to engage with entities that aren't easily observed, labelled or indicated, which we decided to call, for the purposes of our diagramming workshop, "difficult-to-observables." But such entities precisely seem to play an important role when it comes to understanding AI in the street, which indeed is often invisible. Methods of actor-network mapping are also often presentist, pushing us into a narrow preoccupation with current affairs, with what is salient today (and which may be different from what is most problematic and important to address).
Therefore, in diagramming AI in the street, we proposed the following four principles:
1) Mapping of actors and relations can include “difficult-to-observables"
In creating your mapping of AI in the street, don’t hesitate to include structural phenomena (such as global infrastructures, power, inequality, information asymmetry, and so on).
2) Adopt a standpoint in creating your diagram of AI in the street
Rather than assuming a view from nowhere on AI in the street, consider from which perspective you are observing or sensing AI in the street, such as a cyclist, or a pigeon. (To include standpoints in this way will enable us to consider the contested realities of AI in the street).
3) A broad approach to temporality: the expansive present
Adopt an expansive view on what is happening with AI in your street. Whether something has happened or will happen is not always the pertinent question, so is what may or may not be happening. We encourage the inclusion in your diagram of semi-existent entities, and potentialities)
4) Treating data as material / medium
Data does not only have representational value and nor does it only serve to elicit views (provocation or elicitation) Data provides raw material for artistic expression, and not only material for interpretation).
We thus proposed to use exploratory diagramming as a method to map the actors and relations implicated in AI in the street. Unlike traditional diagramming, which represents a specific situation in terms of time, location and scale, our approach challenged the notion of faithful representation in a few ways. Through the inclusion of entities that are difficult to observe and using the temporal frame of the “expansive present,” we recognise that the observation of AI in the street is a contested relationship. By ensuring the diagram presents AI in the street from a specific standpoint, we ensure that experience is included in some shape or form. Asking participants to create keys for their mapping, in the form of symbols, lines and colours signifying important elements, makes the diagram communicable – the key can guide the reader of the diagram through the ontological complexities set out. This process of diagramming produces a graphic form of reasoning where creativity in the form of speculation, narrative, approximation and incompleteness is encouraged.
The Diagramming Exercise
At the start of the exercise, participants were provided with the following materials:
Whiteboards & whiteboard pens
Transparent sheets & permanent markers
Collage images of street entities (people, vehicles, technology, flora, furniture)
For participants joining online, the collaborative drawing platform Figma was used
They were instructed to create a representation of these chosen AI in the street, and to keep in mind that the diagram should be able to communicate AI in the street to someone who is not familiar with their chosen case.
To this end, participants were instructed to begin the mapping exercise by creating a narrative description of an actual or imagined situation in which AI, based on their example, and including moments of observing and not observing AI, are present or felt in the street.
On a template drawing of a street, street entities were added. On top of that, transparent sheets were drawn on and layered so that each layer represented elements of particular cases of AI in the street.
Participants were also asked to consider in which register AI and the relations with AI unfolded. We suggested the following four registers – or modalities - could be in play:
Sensory
Unobservable
Power / Knowledge
Psychic / Potentialities
Our workshop generated a set of wonderful diagrams representing encounters of AI in the street which participants were especially concerned with:
Conclusion
In our discussions, we reflected on how the workshop exercise helped us develop our thinking for the wider project:
There is a temptation in how “AI in the street” has been set up to turn to the street to demonstrate failure, disruption, friction – that is to falsify the utopia of “responsible AI” implied by this promotional and promissory discourse. However, the experience of AI “working well” is just as important – this we equally want and need to specify as lived experience (not just friction).
The success of a particular AI solution must be seen in the context of cascading failures (low walkability, no public transport), a negative loop in which a degraded environment makes AI an attractive solution.
We may need to consider bureaucratic infrastructures as key to the evaluation of AI in the street: given this infrastructure of tick boxes and testimonials it is always possible to demonstrate success. See: The Smartness Mandate by Orit Halpern.
Attending to conspiracy is also important methodologically speaking. Conspiracy has been defined as an undisciplined pursuit of connection, one that does not know when to stop, and which records conjecture as connection. Empirical approaches to social-technical mapping have been defined over against conspiracy in the past: In STS the mapping of socio-technical assemblages is often limited to empirically traceable entities (and this especially in ANT—Actor Network Theory), this doesn’t work for AI in the street. AI features as an absent/present in the street, in some ways it is always “somewhere else” - from a technical perspective, machine learning operates on the data base, not the street.
These observations and questions are now informing the various street-based interventions that are being prepared by project teams in Edinburgh, Coventry, London, Cambridge and Logan and will take place during the coming months.
AI in the street is funded under the AHRC BRAID programme (Bridging Divides in Responsible AI). BRAID is a 3-year national research programme funded by the UKRI Arts and Humanities Research Council (AHRC), led by the University of Edinburgh in partnership with the Ada Lovelace Institute and the BBC.