Artificial Intelligence and Public Administration. Four key words: productivity, discretion, impartiality and territorialisation

The rapid development of solutions based on interactive large language models (LLMS) such as ChatGPT and similar, and therefore on applications based on artificial intelligence (AI), requires equally rapid action by public-sector organisations. The use of these types of technological solutions requires careful reflection so that what has been called ‘algorithmic governance’ and also the ‘third wave of digital era governance’ (Mejer et al. 2021; Dunleavy & Margetts 2023) realise, at least in part, the expectations placed on these technological innovations. 

That is why I believe it is important to address critically, through the use of four key words, some of the issues that characterise the debate on the introduction of AI in public administrations: productivity, discretion, impartiality and territorialisation.

Data lakes, chatbots, and machine learning provide the opportunity to improve productivity 

The first key word is the question of productivity. While the introduction of digitalisation has not always led to efficiency improvements in the public sector (e.g. in the judiciary, where the results have so far been insignificant, although some outcomes have been recorded in terms of the quality of work processes), high expectations are being placed on applications based on Data Science and AI. Despite some positions warning of excessive optimism (e.g. Acemoglu and Johnson 2023), expectations for a number of sectors appear to be based on sound evidence. First, significant improvements could be achieved in all repetitive activities, such as email and certified mail management, or as support for decisions regarding the type of service best suited to the needs of users. For example, INPS already uses machine learning to sort messages received from citizens to the various departments, or to associate the most consistent job activation paths with an applicant’s profile. Chatbots can search for documents in a huge amount of data stored in a variety of forms, whether a query has been raised by a public official or a citizen. Another case is the data lake-based document search systems, through which semantic enrichment and machine learning systems allow the retrieval of any type of material quickly and with greater precision than was possible with previous methods; such experimentation is taking place in the judicial world; consider also the biometric recognition systems (e.g. facial recognition) used at airports, but not only (see Leonardi and Boscaro 2024).

AI offers solutions to increase uniformity levels and avoid discretion 

The second key word concerns the issue of discretion. The debate on the difficulties of various public interventions in achieving consistent implementation is well known, due to the wide discretion of so-called ‘street-level bureaucracies’. This latitude is necessary for tailoring services to the needs of beneficiaries and adapting them to different contexts. However, beyond the opportunistic behaviour of bureaucracies highlighted in the literature, what often emerges are the significant differences in the choices made by public operators despite the presence of similar characteristics of the cases to be judged; these divergences characterise the same decision-makers in relation to cases dealt with at later times, or emerge from comparisons between different operators (see the concept of ‘noise’ attributed by Kahneman (Kahneman et al. 2021) to these situations of excessive disparity). AI can offer solutions capable of increasing uniformity levels through algorithmic decision-making (ADM) applications, based on algorithms that can learn and improve their outcomes over time. Think of the applications of so-called predictive justice, their use in active employment policies to associate the characteristics of the unemployed with training and employment opportunities, medical diagnoses (including predictive diagnosis), tax coherence analysis (ISA Fiscal Reliability Index), etc. (See  Gillingham et al. 2024; Margetts et al. 2024). 

There is a need for transparency in the design of algorithms and on the results of applications

The rapid encroachment of AI into decision-making processes relating to public services and policies  immediately raises the problem of the third key word, that of impartiality. First, as has been widely discussed in literature, in newspapers and on social networks, algorithmic decision support solutions have, on several occasions, resulted in discriminatory outcomes or limitations on privacy protection; this is true, for example, in cases of the use of applications by law enforcement and the justice system in the United States, which have highlighted the need for transparency in algorithm design and application results (see the issue of the transition from discretion of bureaucracies to automated discretion by Zours et. al 2020). A second dimension concerns the ways in which higher productivity in the public (and private) sectors can be pursued; if, as has been mentioned on many occasions, the development of AI were to lead to the expulsion of workers and professionals from the labour market and their replacement by technology, the result would be increased inequality. Whereas, on the contrary, the objective should be to identify higher-value-added tasks that can increase marginal productivity and contribute to progress effects (Acemoglu & Johnson 2023). Finally, a third dimension concerns the problem of access to new AI solutions and the reduction of the digital divide and digital illiteracy, which still affect different categories of citizens and businesses, as well as territories.

New governance models are needed for the balanced dissemination of innovations

All of these factors introduce the fourth key word, territorialisation, which addresses public sector AI application development policies; one of the features of the evolution of digital technologies is the increase in the need for vertical integration among public administrations, with greater relevance assumed by the state level (think public platforms, the cloud, cybersecurity), but without losing the importance of local governments and citizen-facing services (see Dunleavy & Margetts 2023). The development of digitalisation requires new governance models to ensure a balanced diffusion of the innovations under way, also to allow less-equipped administrations (in particular small municipalities) to be able to lock on to the evolution already characterising many large and medium-sized cities (see Giulio e Vecchi 2023). In this direction, the adoption of programmes based on the idea of territorial development and not only of certain specific administrations becomes decisive, both in terms of accessibility to new technologies and of contributing to levels of wellbeing and socio-economic cohesion.

The development of AI applications requires a “pro-human” orientation

The integrated reading of the phenomena linked to these four key words leads to some conclusive considerations. First, the need for “pro-human” guidance that the development and adoption of AI applications should pursue, so that they are used for purposes complementary to individuals, workers, and citizens alike (Acemoglu & Johnson 2023).

Finally, a decisive factor in the introduction and use of AI systems in the public sector is lifelong learning. The literature on the subject consistently points to the fact that the delay in developing digitalisation in public organisations depends in many cases on the difficulty staff have with basic digital skills, combined with an understanding of the appropriate use of applications. In view of the challenges related to the algorithmic solutions mentioned above, interventions on administrative capacities in general terms and on the strengthening of internal and in-house specialist skills become decisive. 

We tend to underestimate how strategic the proper functioning of the public sector is to accompany the economic, social and cultural development of the territories and, more generally, of the country. For this reason, POLIMI GSoM has designed the Executive Master in Public Management and the Online Certification Program in Change Management and Digital Transformation in Italian Public Administrations, so as to seize the opportunities that new technologies offer in relaunching the State administrative machine and the innovations that will be implemented thanks to the National Recovery and Resilience Plan (PNRR).  

 References

Acemoglu D. & S. Johnson. 2023. "Rebalancing AI." International Monetary Fund - Finance and Development, December, pp. 26-29.

By Giulio M. & G. Vecchi. 2023. "How “institutionalization” can work. Structuring governance for digital transformation in Italy." Review of Policy Research 30:406-432, DOI: 10.1111/ropr.12488.

Dunleavy P. & H. Margetts. 2023. “Data science, artificial intelligence and the third wave of digital era governance.” Public Policy & Administration, https://doi.org/10.1177/095207672311987.

Gillingham C., J. Morley & L. Floridi.  2024. “The Effects of AI on Street-Level Bureaucracy: A Scoping Review.” Centre for Digital Ethics (CEDE) Research Paper, available at SSRN: https://ssrn.com/abstract=4823175.

Kahneman D., O. Sibony & C. Sunstein. 2021. Noise. A Flaw in Human Judgment. Turin: UTET.

Leonardi M. and A. Boscaro. “AI IN PA.” Il Foglio, edition of 27 September 2024, 

Margetts H., C. Dorobantu & J. Bright. 2024. “How to build Progressive Public Services with Data Science and Artificial Intelligence.” The Political Quarterly, https://onlinelibrary.wiley.com/doi/full/10.1111/1467-923X.13448.

Meijer A., L. Lorenz & M. Wessels. 2021. “Algorithmization of Bureaucratic organizations: Using a practice Lens to Study how Context Shapes Predictive Policing Systems.”  Public Administration Review, 81(5): 837– 846. 

Zouridis S., M. van Eck & M. Bovens. 2020. “Automated Discretion.” In T. Evans & P. Hupe (eds.), Discretion and the Quest for Controlled Freedom. Cham: Palgrave, pp. 313-329.