Assessment Instruments to Digital Ones: Mapping the Concerns

Abstract

Digital RAIs raise significant concerns, including opacity and discrimination. However, limited to no knowledge exists on the transition from analogue to digital RAIs and on reasons for non-adoption of technology. To investigate this, we reviewed newspaper archives for debates on RAIs and conducted interviews with high-ranking oficials from Moldova. The study yielded considerations for the nonadoption of RAIs and the (non-)transition from analogue to digital RAIs. It highlighted a gap: The lack of a unified theory for technology non-adoption, which can be addressed by drawing insights from innovation resistance and adoption theories.

-

CEUR

ceur-ws.org RAIs.

Specifically, six out of ten interviewees emphasized limited financial resources , with one of them making a comparison to Germany. Five interviewees pointed to the lack of human resources, the scarcity of research centers, and the importance of continued professional education to keep up with new developments. Four interviewees emphasized institutional trust, noting that the judicial role is seen as a human domain and transferring it to algorithms could cause Belgium mistrust. Four interviewees highlighted the issue of individualization versus generalization, arguing that digital RAIs oversimplify complex cases, leading to inadequate decisions. Five interviewees focused on algorithm fairness, questioning whether RAIs can support rehabilitation and prevention of wrongdoers efectively. Linked to this, two interviewees expressed distrust in RAIs’ objectivity, citing the lack of empathy in algorithmic assessments. Three interviewees discussed the fear of being displaced or replaced by technology, citing concerns that RAIs could potentially undermine the authority of legal professionals.

With regard to values, one interviewee pointed out that, similar to European countries, Moldova has been hesitant to adopt algorithmic risk assessment due to cultural views on crime, which is seen as stemming not only from personal attitudes but also from social context. Within this context, a theme brought forward by five interviewees is modernity, emphasizing that the cultural readiness to adopt new technologies is central to the development of RAIs. Furthermore, four interviewees highlighted concerns that these tools would not be applied adequately due to a lack of well-regulated procedures for eficient implementation and issues of corruption. Turning to legal issues, one interviewee noted that current legislation does not permit the use of RAIs at the execution phase, while another raised concerns about potential conflicts with Moldovan law, which mandates consideration of all individual circumstances. Finally, three interviewees cited competing priorities such as energy and health issues, alongside the crucial role of lobbying, as significant constraints for the implementation of RAIs.

Based on the above analysis, among the criticisms voiced against digital RAIs [ 3 ], only the issue of model performance has been addressed in a somewhat direct way. Furthermore, when relating our research to the theory of acceptance and utilisation of technology, our findings suggest that three of the four moderating factors from UTAUT could play a role in explaining non-adoption: age, experience, and voluntariness. Yet our findings introduce two factors that provide nuances to UTAUT categories: distrust of the technology being of poor quality and lack of capabilities to manage the technology. Finally, interviewees’ concerns underscore a gap in UTAUT, which can be addressed by integrating insights from innovation resistance and adoption theories. These insights are relevant to policymakers, legislators, researchers, practitioners, and stakeholders interested in adopting technologies like RAIs. We refer readers to supplementary materials under https://shorturl.at/aqwR0 concerning the keywords used for the archive work, and separately the details of the interviews conducted.

References

[1] C. C. Wolverton , R. Cenfetelli , An exploration of the drivers of non-adoption behavior: A discriminant analysis approach , ACM SIGMIS Database: the DATABASE for Advances in Information Systems 50 ( 2019 ) 38 - 65 .

[2] E. Papagiannidis , I. M. Enholm , C. Dremel , P. Mikalef , J. Krogstie , Toward ai governance: Identifying best practices and potential barriers and outcomes , Information Systems Frontiers 25 ( 2023 ) 123 - 141 .

[3] J. Dressel , H. Farid , The accuracy, fairness, and limits of predicting recidivism , Science advances 4 ( 2018 ) eaao5580 .