Tekoälyn yhteiskehittäminen julkisella sektorilla

Abstract

Artificial intelligence co-creation in the public sector
 This article assesses the steps of AI co-creation in the public innovation ecosystem. However, in order to understand AI in the right context, two aims need to be fulfilled first. The limits and possibilities of AI in public administration are assessed by examining the contributions of AI research, phenomenology and philosopher Hubert Dreyfus. According to the article, the simpler the rules, the better for AI. Therefore, hierarchical and bureaucratic structures are best for AI application. However, bureaucratic structures are poor in innovating and exploration – AI innovations included. After this, co-creation is introduced as a way of innovating AI in the public sector. For identifying new opportunities for co-production, different forms of cross-sectoral and citizen involvement are needed.

Bureaucratic routines and error management in algorithmic systems

“…However, in addition to these studies, scholars have recently started to explore how the longstanding literature on earlier decision-making mechanisms could be applied to inform the design and evaluation of algorithmic systems. These works argue that important insights could be learned by viewing algorithms as analogous to more traditional bureaucratic decision processes (Alkhatib & Bernstein 2019;Autioniemi 2020;Alkhatib 2021;Pääkkönen et al 2020).…”

“…The starting point of this approach is that algorithms are rigorously prespecified sets of rules for information processing, which at the time of execution cannot take into account unanticipated or situation-dependent information that might be relevant for decisions (e.g. Alkhatib & Bernstein 2019;Autioniemi 2020). This irreflexivity of algorithms becomes especially problematic in complex public sector contexts, where the discretionary evaluation of particular cases has traditionally formed the foundation for policy application (Alkhatib & Bernstein 2019).…”

“…At best, the operation of algorithms can be corrected after decisions, if information about their errors becomes available. Due to irreflexivity, Autioniemi (2020) has argued that rule-based artificial intelligence applications in public administration should be primarily limited to routine tasks, where decisions follow from unambiguous criteria (see also Mehr 2017). For the same reason, Alkhatib (2021) argues that machine learning applications with implications for large numbers of people should be abandoned, because their potential for adverse consequences is high.…”

“…Recent discussions of automated decisionmaking have drawn on literature in organizational and administrative studies to analyze the role and shortcomings of algorithms. Rule-based algorithms in public administration have been likened to the strictly circumscribed decision processes of classical bureaucracies, which cannot deal with heterogeneous, ambiguous, or frequently shifting decision-making circumstances (Autioniemi 2020). Similarly to bureaucratic processes, algorithms have been argued to constitute irreflexive mechanisms for decisionmaking, in that at the time of their execution, they cannot accommodate unanticipated or situation-dependent information about the processed cases (see Alkhatib & Bernstein 2019).…”

“…I use the term "algorithm" to encompass both the relatively simple rule-governed mechanisms used to automate tasks in highly regulated contexts such as tax administration, as well as the more sophisticated statistical approaches to data modeling that typically underlie artificial intelligence applications (cf. Autioniemi 2020). By "errors", I mean any results of decision processes, which are unintentionally unfair, systematically biased, or inconsistent with the aims of decision-making.…”