Köbis et al. (2022) – The promise and perils of using artificial intelligence to fight corruption

Question 1

Abstract

Answer 1

Corruption presents one of the biggest challenges of our time, and much hope is placed in
Artificial Intelligence (AI) to combat it. While the growing number of AI-based anti-corruption tools
(AI-ACT) have been summarized, a critical examination of their promises and perils is lacking.
Here, we argue that the success of AI-ACT strongly depends on whether they are implemented
top-down (by governments) or bottom-up (by citizens, NGOs, or journalists). Top-down use of AIACT can consolidate power structures and thereby pose new corruption risks. Bottom-up use of AI-ACT has the potential to provide unprecedented means for the citizenry to keep their
government and bureaucratic officials in check. We outline the societal and technical challenges
that need to be overcome to harness the potential for AI to fight corruption.

Question 2

Advancements in the field of A

Answer 2

Systems that display intelligent behavior by analyzing
their environment and taking actions, with some degree of autonomy, to achieve specific
goals

Question 3

AI-based anti-corruption tools (AI-ACT) are appealing for fighting public corruption. 3 key advantages over traditional anti-corruption efforts:

Answer 3

1. AI systems such as machine-learning can be imbued with autonomous learning abilities
   - AI systems can independently execute various tasks previously reserved to human actors
2. AI can analyze data sets of unprecedented size, thanks to computing power
   - This ability plays a crucial role in keeping track of newly emerging complex corruption
   schemes
   - AI tools can also sift through large data leaks such as the Pandora Papers to unveil
   corrupt patterns
3. AI is impartial
   - Human decision-makers often face conflicts of interests
   - Algorithms pursue no self-serving interests and process information in a disinterested way
   - They make consistent decisions, unaffected by time pressure or fatigue
   - In contexts where corruption is widespread, AI use might be a solution

Question 4

Distinction between Public- and Private Corruption Types:

Answer 4

Private corruption abuses of power not entrusted within the pubic sector, such as embezzlement by managers or bribe payments in business-to-business transactions

• Public corruption à abuses of entrusted power for private gains in the public sector,
  like public corruption ranges from heads of states embezzling public funds to lowerranking public officials like traffic police officers requesting bribes

Question 5

There are three unique challenges that exist for AI-ACT:

Answer 5

1. Data challenge obtaining valid and reliable data to establish a ground truth
   - Data sources typically suffer from biases
   - Documenting and eventually reducing such biases requires an observable ground
   truth, which is difficult to establish for corruption
2. Algorithm challenge à choices made by intelligent algorithms often have far0reaching,
   value-laden consequences
   - AI systems categorize cases into true vs. false
   - Avoiding backlash against AI-ACT requires a balancing act of minimizing falsepositive vs. false-negative error rates

Question 6

AI for top-down vs. bottom-up and its different effects on power structures

Answer 6

Using AI-ACT top-down can lead to a consolidation of power, which introduces new
(corruption) risks
- Power lies with those who have data and code
* A characteristic of top-down AI-ACT is the technical infrastructure of access and
aggregation of (sensitive) data combined with powerful algorithms enabling
unprecedented surveillance and control
* Access and aggregation of (private) databases facilitate the pursuit of corruption à this is
true for most top-down AI-ACT
* The concentration of power tends to breed power abuses
* Big brother effects à governments use AI tools to monitor and weave out political
opposition
* One of the biggest challenges for anti-corruption efforts in corrupt contexts is that those in
power abusing it for private gains have little incentive to change the power structure and
reduce corruption

Question 7

Are AI-ACT in bottom-up efforts flipping the script?

Answer 7

Bottom-up approaches seek to reduce corrupt practices by analyzing the given sociocultural context and support existing efforts by civil society organizations, NGOs, and
investigative journalists
- Enabling protest and other forms of collective action are crucial for democratic
regimes to emerge, and corruption to diminish
* Bottom-up AI-ACT allows citizens to organize better and scrutinize their government à
instead of the government taking the role of a big brother, AI-ACT used in bottom-up
efforts can allow the public to turn into watchdogs, keeping the government in check

Question 8

How to increase the success of bottom-up AI-ACT

Answer 8

1. Bottom-up AI-ACT needs data to fuel the algorithms
   * Data traces à digital technologies like smartphone apps or sensors embedded in
   people’s daily lives collect and store traces of human behavior
   - Browser history, social media communication, geospatial data, etc.
2. To foster and sustain collective action
   * Successful collective action requires the mobilization and sustained engagement of citizens
   * Transparency on paper needs to be turned into action to advance accountability, as transparency without accountability is like the “sound of one hand clapping”
   * AI tools could help transparency efforts by facilitating the reporting of corruption cases
   - AI-based efforts could draw on chatbots or voice-bots that ask about the crucial aspects of the case
   * Dedicated motivator à someone or something who keeps others engaged
   * The success of bottom-up AI-AXT depends on the socio-economic contextL: digital
   collective action requires smartphones, internet access and technical skills (which not
   everyone has access to)

Question 9

Future scenario: coupling AI-ACT with other technologies

Answer 9

• Data transparency can pose a risk to individual privacy
• Unmasked data leaks, neglected privacy concerns, etc.
• Distributed Ledger Technologies (DLT) à data storage systems that use peer-to-peer
  networks of independent nodes
• Every network node stores an identical copy of the database
• Nodes independently validate transactions on the network through an algorithmic
  consensus mechanism
• These transactions are then stored in a timestamped chain of blocks à blockchain