Public encounters now take place across particular combinations of multiple channels, in what is a multi-channel environment (Lindgren, I., & Madsen 2022). Although some in-person work persists, swathes of it has been decommissioned following rounds of digital transformation and disruption from the Covid-19 pandemic. Any worries about implication for trust and empathy have been crowded out by the promise of convenience and cost-effectiveness, as public services adapt to expectations of smartphone consumers (Jeffares 2021). Until recently most AI based channels in the public service have been found wanting, yet recent advances in generative AI seems like a leap forward for both sides of the public encounter. Building on Hancock et al (2020) we might call this genAI-MC, or generative AI mediated communication. Such a possibility is: “Interpersonal communication that is not simply transmitted by technology, but modified, augmented, or even generated by a computational agent to achieve communication goals” (Hancock et al 2020:20). Whereas self-service is human to computer (H2C) interaction, here we have the possibility of both frontline public servant and citizen/consumer communicating where the messages both ways are in some way modified, augmented or generated by generative AIs. This goes beyond ideas of a digital public encounter, which tends to be reserved for remote interaction across digital channels.

Three most important dimensions are goal, magnitude and autonomy. That is, what is the optimization goal the agent is looking to the AI for: advice, persuasion, charm, trustworthiness, diplomacy, leg work? How much is the AI enacting change on the message: minor edits in tone vs much or all of the content? And third, and relatedly, how much supervision does the human agent have over what is sent? Through analysis of national data, adoption varies considerably. Some public organisations in England and Wales are using CRMs with gen-AI capabilities, others are adopting the likes of Microsoft Co-pilot or Google Gemini into their email applications, spreadsheets word processing. For others it is a matter of shadow IT, where public servants covertly use generative AI to help them cope with demands of their role. Similarly local citizens will increasingly be using the likes of ChatGPT to craft the perfect complaint letter or persuasive letters to planning officials. It is relatively early days. After exploring what literature we have on the goal, magnitude and autonomy of generative AI we seek discussion with public servants across a diversity of local governments. Sampling is based on two key factors: the degree to which their organisation has adopted generative AI in policy and practice (such as the development of guidelines or genAI software) and second across a range of citizen voice job roles from those who handle service requests, to complaints, handling reviews and requests for information disclosures (FOI). This work is ongoing, so, if accepted, results will be updated in this abstract closer to the conference.

High administrative burdens push citizens and officials into relying on brokerage to simplify or bypass bureaucratic procedures (Lotta & Marques, 2020; Peeters, 2020). Still understudied in public administration literature, Brazilian despachantes are private street-level brokers who intermediate public services for a fee. They are strongly institutionalised, both as actors in public service systems and as a recognised profession (Bonelli, 2017; Brasil, 2021). Despachantes’ presence is strong even in basic public service provision, such as the welfare cash-based benefit systems provided by the Brazilian National Social Security Institute (Instituto Nacional do Seguro Social – INSS).

To reduce administrative burden, INSS implemented digital self-services, provided mainly through a cell phone app (‘Meu INSS’), which became the main channel for interaction with the agency. The strengthening of remote service channels could be expected to reduce the need for intermediaries, both due to lower administrative burden, and to the weakening of connections between despachantes and street-level bureaucrats, the latter taken from behind the counter to ‘decision factories’ (Peeters, 2023). However, as elsewhere, digital self-services can create new burdens (Lloyd & Wivaldo, 2019; Rydén & De Andrade, 2023), increasing the need for intermediation.

This research asks whether the introduction of digital self-services reduced street-level brokerage of INSS welfare services in the Brazilian State of São Paulo. To answer the question, we rely, first, on the quantitative analysis of longitudinal INSS administrative data concerning benefit application channels (n = 415.373, 04.2019–03.2020). Second, we employ thematic analysis of open-ended survey questions concerning despachantes and digital self-services, answered by 167 INSS officials working in the State of São Paulo. São Paulo shows the strongest presence of INSS services intermediation among other Brazilian states.

Findings suggest that the introduction of digital self-services, instead of keeping street-level brokerage at bay, often made them invisible to the agency’s radar. The study thus provides relevant insights for administrative burden theories and proposes street-level brokers to be accounted for as actors in an unintended hybrid service governance system.

There has been a sharp increase in public organizations' interest in and use of AI. In particular, those involved in the police have a keen interest in using AI-based recommendations in investigations, internal processes, and service delivery. The promise is for this to enhance and streamline investigative and administrative work. Yet the opaque and complex nature of AI recommendations is heavily criticized for worsening extant biases in decision-making. The loosely connected confederacy of parties working on Explainable Artificial Intelligence (XAI) is poised to address these issues. XAI models developed by computer scientists are designed to offer non-technical users transparent explanations of why or how an AI recommendation was reached. XAI is hypothesized to decrease the risk of undesirable, biased decisions based on AI recommendations. The XAI movement aligns with calls for increased explainability, and algorithmic transparency aimed at understanding AI outcomes. Among other things, initial user studies found that XAI and increased explainability generally had a positive impact on user trust in AI recommendations.

In this paper, we argue that increasing trust should not be a goal in and of itself: undue trust in AI recommendations can cause them to be unduly integrated into a final decision. Additionally, though understanding the attitude towards AI is valuable, we argue only as far as it helps us understand the outcome of human-AI decision-making. Here, we test whether XAI contributes to more conscious deliberations on AI recommendations, and as such contributes to appropriate reliance as seen in decision outcomes. In other words, XAI should help police officials rely on AI recommendations when they are likely to be correct, but they should remain vigilant and thus reject AI recommendations when they are likely to be incorrect. Such vigilance ought not be limited to initial interactions, but ideally remain in subsequent decisions.

We propose a series of survey experiments with which we have a threefold goal; first to test the effect of error ordering on appropriate reliance, Second to test the effect of explanations on decision-making and third to do so in an environment representing extended use of AI recommendations because of repeated decision-making. We do so using highly realistic tasks for the police context.

With the proposed study, we aim to contribute to our understanding of government decision-making involving AI. We extend transparency research by focusing on the outcome of human-AI decision-making as opposed to the experience of a decision recipient. We also aim to improve our understanding of how human-AI decisions are moderated by contextual factors such as repeat decision-making and provide a novel as well as flexible experimental strategy for examining such questions.

The digital transformation of the public sector has spurred significant changes in administrative

processes and service delivery, aiming to enhance accessibility and efficiency. However, this

shift brings forth both opportunities and challenges, particularly concerning the interaction

between citizens and the state. Furthermore, the literature reveals a fragmented understanding

of how digitalization influences administrative burdens, especially among marginalized groups.

Addressing this gap, the paper conducts a systematic literature review to identify key factors

that shape these burdens in a digitalized context. Consequently, the paper addresses the

following research question: What influences administrative burdens in digital public

administration? By synthesizing theoretical considerations and existing empirical evidence,

this review highlights the complex interplay between digitalization and administrative burdens,

offering insights for future research, policymakers and practitioners in public administration.

The findings show that administrative burdens in digital administration are influenced by

characteristics of the administration, digitalization, and citizens. Key factors include

administrative assistance, information infrastructure, and citizens' infrastructure capital. Issues

such as biased data, lack of caseworker assignment, and inadequate resources exacerbate these burdens, while social capital and administrative cooperation can alleviate them

Artificial intelligence (AI) holds great promise for public administration as it redefines the ways in which the public sector creates policies and services (Berryhill et al., 2019). The implementation of AI in the public sector has become increasingly relevant in recent years, as evidenced by the growing number of research papers on the topic (Valle-Cruz et al., 2024) and empirical experiences of AI adoption in public entities (Maragno et al., 2023). Governments are using artificial intelligence (AI) to enhance policy design (Bellini et al., 2022), make better strategic decisions (Young et al., 2019), improve citizen relationships (Reis & Melao, 2023), foster citizen’s trust in governments (Dwivedi et al., 2019), and improve the detection, prediction, and simulation of public services (Margetts & Dorobantu, 2019). Additionally, AI is being used to enhance operational efficiency (Ubaldi et al., 2019) and improve the quality of public services (Wirtz et al., 2019; Benouachane, 2022; van Noordt & Misuraca, 2022; Yigitcanlar et al., 2021).

Although the potential benefits are significant, achieving them is not an easy task. There are many risks and limitations that must be considered, such as data availability, interoperability, standards, privacy, security, and ethical issues (Ubaldi et al., 2019). Furthermore, the impact of AI systems on public governance remains unclear for many governments due to their limited understanding of the multifaceted implications of AI usage (Zuiderwijk et al., 2021). As governments are generally risk-averse and not open to discussing failure (Albury, 2005; Brown & Osborne, 2013) for both political and reputational reasons (Rhodes, 2011), they tend to be late in adopting AI technologies compared to the private sector (Desouza et al., 2020). However, they are required to adopt AI tools ensuring a balance between AI potentialities and AI risks by adopting a responsible, trustworthy, and human-centric approach to the use of AI in the public sector (OECD/CAF, 2022). This leads to the need of the development of effective AI strategies, which requires an understanding of the attitudes towards AI adoption. Although citizens' attitudes towards this issue have been analysed (O'Shaughnessy et al., 2023), and a number of scales against which these can be determined have been developed and tested (Kaya et.al., 2024, Bergdahl et. al., 2023; Grassini, 2023; Schepman & Rodway, 2020), there has been no examination of the stance taken towards AI from the perspective of public administration.

The purpose of this paper is to investigate government attitudes towards AI adoption through a comparative qualitative analysis of national and local strategy documents. Based on a coding scheme derived from the literature, the goal of the paper is to explore predominant attitudes towards AI embedded in the national strategies of Austria and Spain, and to determine whether these are reflected – indeed, transposed – in the local government AI/digital strategies of the cities of Vienna and Madrid. A scoping review of literature on smart cities, and of the scales developed to measure general attitudes towards AI, will underpin the analysis. In this way, the paper will seek to identify differences between theory and practice and design a basic theoretical framework that outlines government attitudes towards AI.