Corruption presents one of the biggest societal and political challenges of our time. It undermines efficiencies of public institutions, widens gaps in inequalities and thereby hinders the achievement of the UN sustainable development goals (Heywood, 2018). To counter corruption, recent hopes have been placed in advanced digital technologies. Studies show that distributed ledger technologies (Kossow, 2018), e-procurement processes (Kovalchuk et al., 2019) and data journalism (Köbis & Starke, 2017) can help to detect, prosecute and prevent corruption. Recent developments in artificial intelligence (AI), defined as “systems that display intelligent behaviour by analysing their environment and taking actions – with some degree of autonomy – to achieve specific goals” (European Commission’s High-Level Expert Group on AI, 2019), demonstrate an almost unlimited potential of machine learning technologies to find novel solutions for old problems. For instance, AI-based systems have been used to diagnose specific forms of cancer more efficiently. Against this backdrop, anti-corruption (AC) agencies are lagging in their quest to tap into the potential of AI to fight corruption. A glaring example stems from “Operation Serenata de Amor” (https://serenata.ai/en) that uses machine learning technologies fueled by open government data to detect red flags in the behavior of Brazilian members of Congress. An automated bot then automatically publishes these suspicious cases via social networking sites such as Twitter. However, while collections of anecdotal case studies about projects that use AI in AC exist (Aarvik, 2019), a coherent conceptual framework is lacking. Filling this gap, we introduce such a conceptual framework that outlines how these emerging technologies impact AC efforts along two dimensions. First, AC consists of three procedural steps: (1) As a prerequisite, it relies on detection of corrupt practices, (2) AC requires effective and impartial prosecution of corrupt behavior, and (3) as the ultimate goal of curbing corruption, AC achieves successful prevention of corruption. Second, along these constitutive steps, digital technologies differ in their degree of autonomy. On the one side of the spectrum, controlled technologies store information, which then requires human analysts to detect corrupt practices in the data (human decision making). Somewhat more autonomous, recent technological advances allow algorithms to independently identify suspicious patterns in the data and report them back to human AC agents (human-in-the-loop). Even more autonomous, algorithms can not only independently detect, but also automatically publish suspicious cases - human actors play the role of an auditor who can only influence the process after disclosure (human-on-of-the-loop). Finally, although not yet implemented, future AI technologies could act fully independently without human interference in the procedure (human-out-of-the-loop). These novel technological advances bear immense potential for AC, such as overcoming the collective action problem of AC agencies themselves falling prey to corruption. At the same time these technologies pose new ethical challenges, such as increased risks of privacy breaches and surveillance. Our novel conceptual framework systematically clusters the existing applications, demonstrates their potentials and perils, and identifies existing knowledge gaps in the academic and policy literature on the AI-AC nexus.
