There are two reasons that people ask this question. First, hope – city leaders are looking to make efficiency and budget gains, and turn to the promise of new technology. Second, and following on from that, fear – there are concerns over accountability and public scandals.
There are many different types of AI, and many different types of applications. It can feel like an overwhelming question; the recent AI and Cities report is over 120 pages long.
Instead of trying to catalogue a technical overview of applications of all the different ‘types of AI’, which are ever-evolving, I suggest a functional approach. Thinking about how cities are using AI and how that changes the way the city works can highlight where we need to focus efforts and resources for learning and capacity building.
Two major functions for urban AI
Over the last years, I’ve seen two broad categories of how artificial intelligence is being used in cities: automating existing processes, and data-driven predictions.
Automation means automating a part of existing bureacratic processes or urban services. In this category, there is a logic or a process that already exists, and one part of that chain of events is going to be made faster or more efficient with the assistance of AI. When considering how to apply AI, the starting block is the current system.
Data-driven predictions is a different approach, because the starting point begins elsewhere: with a lot of data. Out of that data, data analysts will derive insights, and based on those insights, the administration designs new bureacratic processes for urban services. Predictive modelling forms a new, data-driven logic in the administration.
Examples will follow; what’s important is that while these two categories may use the same type of AI on a technical level (for instance, they may both use deep learning or image recognition techniques), the way that the AI is embedded within the processes of the city differs.
That’s important to think about because how AI is embedded in the processes of the city changes the types of impacts that AI can have, and it changes the way an organisation adapts to the new use of technology. The shift to data-driven processes changes the role of local expertise and where it is applied (1). Bringing AI on board to the city shifts organisational culture as well.
How AI is embedded in the processes of the city changes the types of impacts that AI can have
Automation must enable people and support their work
The first category of how AI is used in cities is automation. An example of automating existing bureacratic processes using algorithmic systems comes from the city of Barcelona, which has a constructive municipal data strategy for Ethical AI.
In Barcelona’s Social Services center, an algorithm was designed to support case workers in recommending the next resources for a particular case. The existing process is that residents requiring assistance have an interview with a case worker, who then compares the current case to previous cases and looks up suggestions, resources, and next steps to recommend. This is a paperwork heavy process. Based on a corpus of 300,000 interviews, a machine learning algorithm now automates the recommendation process. The idea is that for the caseworker, their job is now easier because they don’t have to go through all the paperwork and check it manually. As a resident, the intake process at social services still looks the same, presumably easier and faster.
Importantly, the caseworker has the ultimate decision. The person who has all of the experience and knowledge of working with and interfacing with people, still has the final say.
Obviously, caseworkers are at the heart of social work. Automation should not automate people, but enable them to make their lives easier. Integrating AI in these types of cases should focus on enabling those people to do their job better, from their perspective.
Data-driven predictions require local knowledge
The second category of the way that algorithmic systems are integrated into cities is data-driven predictions. These tend to be the more sexy-sounding examples, and a lot of the buzz around AI, both positive and negative, derives from these types of functions.
An example comes from the city of Montreal, where there is a significant fire risk to a lot of the houses made of wood. A challenge the city faces is how best to allocate the resources of where fire trucks should be dispatched to be able to respond faster to fires.
Local developers created a machine learning algorithm to predict fire risk as a way to support the firefighting service. The risk prediction is based on historical data around where fires break out, but also connecting that with a database on construction and building materials, and specifically with age-related maintenance issues.
If you were to provide me with a database of all the construction material in the city, I wouldn’t necessarily know what to do with it. What is interesting is that the risk profile it based on people working locally that have an understanding that this is what connects to fire risk. It is local understanding that can connect databases in a relevant and meaningful manner to generate predictive insights.
Predictive applications really emphasise the role and the need for local knowledge; not only for building contextually appropriate connections, but also to ground truth, design, create and check how systems are designed. (2)
AI implementation means organisational culture change
In order to develop technology in line with public values, incorporating artificial intelligence into city processes needs to shift away from a sole focus on the technology itself.
As organisations incorporate algorithmic systems, those organisations change. The social norms change, the ways of working change, people’s responsibilities change – often, whether people are prepared for it or not. (3)
There are structural socioeconomic impacts too, in that the dependence on data intensifies, public agencies increasingly rely on private data providers, and public agencies can become more commodified. (4)
These organisational and structural shifts require a broad perspective on accountability and governance in order to keep the focus firmly on what we want these systems to do and which interests they serve. Valuable change in this direction requires a sense of responsibility and vision.
There are promising avenues on how to address some of these issues all along the AI life cycle; from development, procurement, implementation, testing, impact and decommissioning. Some cities are starting with this work, though there is lots more work to do.
If you’re working through these issues in your organisation, or considering starting to, I’d love to hear from you. Send me an email at [email protected].
You can also hear/see me elaborate on these ideas of how to integrate AI into urban development as part of this panel from the ITU webinar series of Digital transformation for cities and communities. My talk starts at the 19 minute mark.
References
(1) Kitchin, . (2016) The Ethics of Smart Cities and Urban Science. Phil. Trans. R. Soc. A 374, no. 2083. https://doi.org/10.1098/rsta.2016.0115.
(2) Costanza-Chock, S. (2020) Design Justice: Community-Led Practices to Build the Worlds We Need. MIT Press. https://mitpress.mit.edu/9780262043458/
(3) Jameson, S., Taylor, L. and Noorman, M. (2021) Data Governance Clinics: A New Approach to Public-Interest Technology in Cities. SSRN Scholarly Paper. https://doi.org/10.2139/ssrn.3880961.
(4) Baykurt, B. (2022). Algorithmic accountability in U.S. cities: Transparency, impact, and political economy. Big Data & Society, 9(2). https://doi.org/10.1177/20539517221115426