[Paper Review] Artificial Intelligence as an Anti-Corruption Tool (AI-ACT) -- Potentials and Pitfalls for Top-down and Bottom-up Approaches

Nils Köbis, Christopher Starke+1 more|arXiv (Cornell University)|Feb 23, 2021

Ethics and Social Impacts of AISocial Sciences58 references17 citations

TL;DR

This paper proposes a conceptual framework for using Artificial Intelligence as an Anti-Corruption Tool (AI-ACT) in both top-down and bottom-up governance approaches. It analyzes AI-ACT's potentials and pitfalls across data input, algorithmic design, and institutional implementation, emphasizing the need to integrate societal perspectives to ensure ethical and effective anti-corruption outcomes.

ABSTRACT

Corruption continues to be one of the biggest societal challenges of our time. New hope is placed in Artificial Intelligence (AI) to serve as an unbiased anti-corruption agent. Ever more available (open) government data paired with unprecedented performance of such algorithms render AI the next frontier in anti-corruption. Summarizing existing efforts to use AI-based anti-corruption tools (AI-ACT), we introduce a conceptual framework to advance research and policy. It outlines why AI presents a unique tool for top-down and bottom-up anti-corruption approaches. For both approaches, we outline in detail how AI-ACT present different potentials and pitfalls for (a) input data, (b) algorithmic design, and (c) institutional implementation. Finally, we venture a look into the future and flesh out key questions that need to be addressed to develop AI-ACT while considering citizens' views, hence putting "society in the loop".

Motivation & Objective

• To address the persistent global challenge of corruption by evaluating AI as a transformative tool for anti-corruption efforts.
• To examine how AI can support both top-down state-led and bottom-up citizen-driven anti-corruption initiatives.
• To identify key risks and limitations in data quality, algorithmic design, and institutional deployment of AI-ACT.
• To advocate for embedding societal values and citizen perspectives into AI-ACT development, positioning 'society in the loop'.
• To provide a structured conceptual framework for future research and policy design in AI-driven anti-corruption governance.

Proposed method

• Develops a conceptual framework to analyze AI-ACT across three dimensions: input data, algorithmic design, and institutional implementation.
• Compares top-down (state-led) and bottom-up (citizen-led) anti-corruption approaches in the context of AI deployment.
• Examines data sources such as open government data and administrative records as foundational inputs for AI-ACT.
• Analyzes algorithmic design challenges including bias, transparency, and fairness in AI models used for corruption detection.
• Assesses institutional factors such as legal frameworks, oversight mechanisms, and accountability structures for AI-ACT.
• Proposes a participatory approach to AI development that incorporates public values and citizen feedback to ensure legitimacy and trust.

Experimental results

Research questions

• RQ1 How can AI be effectively leveraged in both top-down and bottom-up anti-corruption strategies?
• RQ2 What are the key risks and limitations of AI-ACT in terms of data quality, algorithmic bias, and institutional integration?
• RQ3 How can citizens' perspectives be meaningfully incorporated into the design and deployment of AI-ACT?
• RQ4 What institutional and ethical safeguards are necessary to ensure accountability and transparency in AI-ACT systems?
• RQ5 In what ways do data availability and algorithmic transparency affect the legitimacy and effectiveness of AI-ACT?

Key findings

• AI-ACT presents unique potential for detecting corruption through automated analysis of large-scale government data, especially when data is open and accessible.
• Top-down AI-ACT approaches are more likely to benefit from institutional infrastructure but risk centralization and lack of public trust.
• Bottom-up AI-ACT initiatives can enhance civic engagement and transparency but face challenges in data access and technical capacity.
• Algorithmic bias and lack of transparency in AI models pose significant risks to fairness and legitimacy in anti-corruption applications.
• Institutional implementation of AI-ACT is often hampered by weak oversight, unclear accountability, and insufficient legal frameworks.
• Incorporating societal values and citizen input into AI-ACT development is essential to ensure ethical deployment and long-term public acceptance.