GAVIUS Journal #1: When AI Serves Common Good

Although the concept of a data-driven ‘smart city’ has been around for decades, many municipalities have yet to deploy the fundamental infrastructure required to harness the full potential of data analytics and AI.

The potential of AI for public administrations is manifold. The development of AI technologies is already having an impact on how the public sector works and designs policies to serve citizens and businesses. Applications touch on areas such as health, mobility and security services.

Local, regional and national governments are implementing projects aimed at exploiting AI to better meet the needs of public-service users and workers. Many use cases are also aimed to enhance stewardship and promote a more efficient use of public resources (e.g., increasingly saving the time civil servants spend on user support and administrative tasks so as at to optimize the use of staff capacities and time to attend to all eligible users and increase quality of the assistance provided)[1].

According to a recent report commissioned by the Administrative Conference of the United States to Stanford University and New York University[2], AI tools are already improving agency operations across the full range of governance tasks, including:

• Enforcing regulatory mandates concerning workplace safety, health care, and environmental protection;

• Adjudicating government benefits and privileges, from social aids to intellectual property rights;

• Monitoring and analyzing risks to public health and safety;

• Extracting useable information from the government’s massive data streams, from consumer complaints to weather patterns; and

• Communicating with the public about its rights and obligations as welfare beneficiaries, taxpayers, asylum seekers, and business owners.

Nowadays the AI tools available to governments span the full technical scope of AI techniques, from conventional machine learning to more advanced “deep learning” with natural language and image data. Despite the promising results achieved in many countries[3], European cities are still lagging behind in large scale implementation of AI in public services’ design, delivery and monitoring. Several struggles make still difficult for public administrations to keep pace with the dynamic and fast development of AI[2]. Among those, there is the need for further work in terms of legal structures to support data sharing, usage and integration; improving standards; practitioners training; digitalisation for work practice and users support.

Moreover, when it comes to utilize AI based systems for the adjudication of social benefits, which is the core e-Government domain that GAVIUS aims to address, the public needs to be confident that the algorithms powering machine learning are fair and verifiable. For this reason, robust information governance and explainability of the algorithms are key prerequisites when deploying AI in the government sector.

Many national governments around the world have devoted considerable resources into developing strategies to foster innovation and development in AI technologies based on a widely shared belief that these technologies can and will deliver very significant benefits in terms of enhanced efficiency, productivity and service delivery.

A comprehensive overview of national AI strategies is made available by the European Commission through the AI Watch platform aimed to monitor the implementation of the Coordinated Plan on Artificial Intelligence on the development of AI in the EU and built in close collaboration with the OECD AI Policy Observatory. In parallel to this ongoing effort, the European Commission published its ‘White Paper on Artificial Intelligence: a European approach to excellence and trust’[5] in February 2020.

Gavà and Mataró are two municipalities of the Barcelona Metropolitan Area with a population of 47,057 and 129,661, respectively, and an intermediate population density (Statistical Institute of Catalonia, IDESCAT, 2020).

Both cities have several e-government services and have developed their own Open Data portals indicating their commitment to digital transformation, citizen participation, transparency and willingness to generate economic and social value through the use and reuse of data. The portals currently host respectively 129 datasets in Gavà and 43 in Matarò. The huge number of records stored will be paramount for the machine learning process that is currently under development in the GAVIUS project.

In the same direction, in 2017, the Big Data Gavà project was launched. In order to value the data generated around citizens in the closest administration, i.e. the city council, it seeks to standardize and systematize the core data of the municipal system, avoiding duplications, incorrect or outdated information and, in turn, feed a Big Data system from which to draw dashboards for decision making support.

GAVIUS specifically focuses on social services as these are core pillars of the pivotal role played by municipalities in promoting equality and equal opportunities relating to education, employment, housing, health and care and cultural and leisure activities. In particular, social aids and benefits are crucial to enable inclusion and thus a precondition for a sustainable welfare society and for individuals to be able to live free and independent lives. Moreover, overall, social services represent a substantial part of local administrations’ budgets, which is planned to reach approximately the 9,4% and the 9,66% in 2021, respectively in Gavà and Mataró.

Based on their shared goal to increase efficiency and effectiveness of social policies, Gavà and Matarò have joined forces and been successful in obtaining funding from the Urban Innovative Actions Initiative to implement the GAVIUS project. GAVIUS has been collaboratively designed by the departments of Innovation and Social Services of the two municipalities and the main local entities involved in the delivery of social services thus bringing on board the voice of a diverse and significant group of stakeholders from civil society.

It has been a year and a half since the kick-off of the project. Although the Initiation Phase took longer than planned due to administrative issues to be dealt with, the Managing Urban Authority has been successful in keeping the project on track avoiding that the administrative delay affected the co-design and technical development work. In addition, the COVID-19 pandemic has disrupted the activity flow as it arose when GAVIUS team was defining their strategy to engage citizens and planning their first participatory dynamics. This has obliged to identify alternative modalities and adapt to the restrictions enforced during the three pandemic waves Spain has gone through and partially still in place today.

Co-creation sessions have been carried out via a series of online focus groups, coupling the use of Zoom for videoconferences with specific applications for the development of the participatory dynamics (e.g., Miro and Mural).

At the same time, the legal, ethical, cultural, social and organizational issues requiring specific attention have been addressed. They are being collected in a living document that is developed throughout the project and will be condensed in a Manual at its end.

On the other hand, an important strand of the work has involved the analysis of the types of aids granted by the two city councils and the data formats and sources. Work is also underway on the conceptualization of the technology platform and the harmonization of data which is a prerequisite for the design of AI algorithms.

In spite of the obstacles posed by the COVID-19 restrictions, the GAVIUS consortium has designed a robust stakeholder engagement methodology (see figure below) and it is currently carrying out the activities foreseen in its 3rd Phase “Design Co-Creation”.

Overall, the innovative solution that GAVIUS aims to develop and test is composed by:

• A virtual assistant for municipal workers: it allows municipal technicians to anticipate the needs of citizens and / or improve their relationship / interaction with the administration, adapting services to the needs of the citizenry through their personalization, improving the citizen’s experience while reducing the time spent by technician in repetitive and automatable tasks.

• A virtual assistant for municipal managers and politicians: it will provide access to a dashboard visualizing all the data and the indicators to help them efficiently allocate resources for social services.

• A virtual assistant for citizens: through secure identification mechanisms citizen can autonomously access and perform simple but time-consuming tasks to be granted municipal aids and services.

A “Proposal - Application – Feedback” system is the backbone of the technological platform thus evolving from the traditional reactive model “Application – Feedback” to a proactive one.

Broadly speaking, the process would be:

• The AI module analyses the characteristics of the potential beneficiary and crosses them with the data included in the historical database of aid requests and responses. This way, it will draw up a proposal fitting the needs and characteristics of the potential recipient.

• The citizen is notified of the aid proposal (Proposal).

• If the citizen agrees, he or she formalizes the aid application (Application).

• The aid management process begins. The compliance of the citizen to the grant eligibility criteria are reviewed and validated and the applicant is notified of the decision (Feedback).

The picture below illustrates the GAVIUS “data journey” across the different steps of the planning, implementation and monitoring of a grant campaign.

One of the keystones of the project is the Handbook on the “legal, ethical, cultural, social and organizational limitations of the project”. This handbook aims to analyse ways of making the use of AI in the administration compatible with the full exercise of our fundamental rights.

The legal overarching framework is provided by the General Data Protection Regulation (EU) 2016/679 (GDPR) which makes no direct reference to AI, or similar technologies, though it does refer to automated large-scale data processing, as well as profiling and automated individual decision-making (where AI systems could be framed).

In Europe, the general principles of protection of the rights and freedoms of individuals have been detailed in the guidelines released by the High Level Expert Group on AI[9], according to which the reliability of AI is based on three components that must be met throughout the life cycle of the system: AI must be lawful, ethical, and robust, both from a technical and social point of view.

The Manual aims to integrate the most advanced hypotheses and current top-notch approaches with the objective of supporting the deployment of tools through the GAVIUS project. For instance, the Manual will conceptualise how GAVIUS will integrate the principles of minimisation (i.e., ensure that the data collected is minimised e.g., identifying innovative solutions that avoid unnecessary personal identification to access information of public interest) and privacy by design and by default.