Nine cities in Europe have developed a free to use open-source 'data schema' for algorithm registers in cities. Those cities are Barcelona, Bologna, Brussels Capital Region, Eindhoven, Mannheim, Rotterdam and Sofia, based on the example set by Amsterdam and Helsinki.
Transparency of AI systems is a hot topic. The UK's Centre for Data Ethics and Innovation have updated their Algorithmic Transparency Recording Standard, building on their reports about bias in algorithmic decision-making. We can expect transparency to be discussed as part of the inquiry in the UK as to what good governance of AI looks like. We can also expect an EU database for high-risk AI systems under the EU's proposed EU AI Act (Article 60).
Here we briefly explain the issue of transparency in how public bodies use AI systems, and points to note from the experiences of cities developing algorithmic transparency registers.
How are cities using AI?
The examples in Amsterdam include reporting issues (e.g. rubbish or maintenance problems), illegal holiday rentals, and parking controls. You can view their register here.
However, identifying when and where public authorities are using AI is difficult to establish because the use of AI systems is not always published.
The lack of transparency is an issue. It is one of the limitations cited by studies into public authorities' use of AI (for example, a report on cancelled AI projects in the public sector here). Without transparency, members of the public do not know when and where their data is being use, how it is being used, and how to seek redress if there are problems.
The growing role of transparency registers
To address the current lack of transparency of when and where AI is being used, some public authorities have been developing AI / algorithmic decision-making transparency registers.
The report explaining how the nine European cities have developed them explain an AI register as "an online list of these various algorithms, with a brief explanation of where, how and why they are being used."
Developing such a register is not straightforward. It requires considering questions including: What scope of information is needed when we want to collect data? How can we collect meaningful, comparable data that leads to greater efficiencies in e-governance, rather than inconveniences? And, how do we use and manage data for various applications?
The nine cities have developed the 'data schema' now whilst AI use is in its relevant infancy. Better to agree on common standards now before wider and greater adoption of more advanced AI in public services.
The aim is for the common approach to be shared and copied by other cities who may not have the resources available to invest in their own initiatives otherwise.
What issues arise with transparency registers?
The experiences of the cities who are developing, or have deployed, transparency registers are useful for highlighting topics relevant in many scenarios when deploying AI - within public or private organisations. For example:
- what systems should and should not be included in the register: Pasi Rautio, Service Manager in the strategy department of Helsinki, explained “We have almost 40,000 people working for the city ... and we have lots of projects going on. So, we don’t know if we have registered entirely all our algorithms, and there is also a question over the definition of AI, and what exactly should be included in the registry.”
- the need for specialist advisory teams within the organisation. The report notes: "One way that Helsinki has affected a better approach to this is by creating specific teams within the city administration that work on targeted aspects of the digital transformation, such as AI or data and analytics. In this way, when other colleagues within the city administration start handling something that might involve AI, they are now much more likely to do so in conjunction with the team of specialists, and consequently know that any algorithm is shared in the registry."
- the need for an AI advisory board. Rotterdam's Innovation Officer explained that “we created an external Algorithm Advisory Board that looks at what we’re doing and whether we’re doing it in the right way. The idea is to remain transparent to all of our stakeholders, whether they’re within the administration or external.”
- questioning what information the public needs for transparency. The Smart City and Innovation Officer for Brussels noted the challenges of: "Finding the right balance in providing adequate information about an algorithm without overloading people, and making sure their use is not seen as an additional burden but as a public good, are all items on the minds of administration officials."
complex algorithms in automated systems trained on biased data may transpose bias to groups of citizens. Algorithm registers offer transparency about the development and implementation of algorithms and provide an important safeguard for the responsible use of AI.