Algorithmic Accountability: Designing Governance for Responsible Digital Transformations

AI- og (selvlærende) algoritmer blir i økende grad brukt til å støtte, påskynde og til og med erstatte beslutninger tatt av mennesker på ulike offentlige og private arenaer. Algoritmer tar i økende grad avgjørelser innen aksjehandel og finans, svindeloppdagelse, vitenskapelig oppdagelse, medisinsk diagnostikk og online match-making. Beslutninger tatt av systemer for kunstig intelligens er ofte implisitte og usynlige, og de kan føre til både tilsiktede og utilsiktede konsekvenser. Dette har ført til at systemer for kunstig intelligens i økende grad har blitt utsatt for offentlig granskning og bekymring. Med dette som bakgrunn, er hovedmålet med prosjektet å se fra et forretningsetisk perspektiv hvordan sosiale, kommersielle og politiske aktører på lokal og global skala kan sikre ansvarlighet i algoritmiske beslutningsprosesser. Prosjektet består av en gruppe internasjonale forskere med ekspertise innen jus, internettstudier, informasjonssystemer og ledelsesforskning. I løpet av prosjektet skal det gjennomføres en flermetodisk undersøkelse blant flere interessenter for å utvikle et omfattende rammeverk for å kartlegge bekostninger, ansvar og utfall av algoritmisk handlekraft. For å få til dette skal vi først utvikle et rammeverk for ansvarlig algoritmisk beslutningstaking forankret i litteraturen om legitimitet, deltakelse og inkludering. Videre skal vi systematisk samle inn, kartlegge og sammenligne ulike foretillinger av algoritmisk handlingskraft. Her vektlegges viktigheten av en samstiftelse av algoritmisk handlekraft mellom organisasjoner og deres interessenter. Vi vil også utvikle retningslinjer som skal bistå i å lage ansvarlige algoritmiske beslutningsprosesser, basert på både lett forståelig programmering og forståelig kommunikasjon av beslutningsgrunnlag og datakilder. Til slutt, som en praktisk leveranse, vi vil lage en normativ modell for å evaluere ansvarlighet i algoritmiske beslutningsprosesser ved å undersøke i hvilken grad algoritmene er gjenomsiktige, tilbyr ordentlige kanaler for konflikthåndtering og muliggjør offentlig tilsyn.

By identifying and addressing the information asymmetries between technologists, industry, policymakers, and society at large, the project has provides policymakers and organizations with the necessary knowledge to oversee and implement AI responsibly by introducing tangible best practices, model legislation proposals, and accessible informational resources through interdisciplinary collaboration and engagement with stakeholders. Concurrently, the project promoted the implementation of more adaptable and inclusive managerial and regulatory instruments by analyzing a comprehensive "governance toolbox" that was informed by prior global governance challenges. This expansion of available alternatives—which encompasses ethical codes, technical guidelines, and legal frameworks—has facilitated enhanced dialogue and coordination. Additionally, the project promoted increased collaboration with academia, which in turn facilitated the establishment of forums and networks that facilitate robust discourse on global AI governance. These endeavors collectively fortified research partnerships among institutions and provided a more lucid pathway for the integration of academic insights into industry practices and policy. The project helps to ensure that AI systems are developed, regulated, and adopted with greater accountability and transparency by reducing informational gaps. The foundation for a more ethically sound and resilient AI deployment is established by the introduction of a diverse set of governance mechanisms, which ultimately fosters public trust and enables AI-driven innovations to serve the public good. The establishment of robust academic-policy-industry collaborations also offers sustainable pathways for addressing future AI challenges, thereby contributing to more equitable and inclusive development. Ultimately, these advancements may shape a more transparent and socially beneficial digital ecosystem by inspiring further cross-border regulatory alignments, promoting stronger international consensus on responsible AI practices, and influencing global policy discourse.

Building on ongoing international discussions, and in continuation of the team’s previous work within the SAMANSVAR framework on fair labor on platforms and the gig economy, we want to address the socioeconomic effects of increasing adoption of artificial intelligence, algorithmic management, and smart automated systems, whereby increasing productivity gains for organisations are met with concerns about whether the speed of implementation can be matched with necessary levels of accountability and oversight. Our research shall result in an unifying framework for algorithmic accountability for algorithmic management which shall serve as a basis for organizations, regulators, and social communities to take actionable steps towards ensuring accountable algorithmic decision-making processes. To this end, we will first draw a framework for accountable algorithmic decision-making grounded in the literature on legitimacy, participation, and inclusion. Second, we will systematically collect, map, and compare varying notions of algorithmic agency. Here, we will set a particular emphasis on the importance of a ‘co-constitution’ of algorithmic agency between organizations and stakeholders. Third, we will develop actionable guidance towards creating accountable algorithmic decision-making, based on both explainable programming and comprehensible communication of decision-making rationales and data sources. As a practical deliverable, we shall create a normative model for evaluating accountability in algorithmic decision-making processes, examining to what extent the algorithms are transparent, provide proper dispute channels, and enable public oversight. Finally, we will develop a user-friendly accountability enhancing tool, the ‘Framework for Algorithmic Accountability’. The framework will be tested on real-world settings of algorithmic decision-making processes and can be utilised by researchers, activists or lay Internet users to challenge algorithmic systems.