From open data to artificial intelligence: the next frontier in anti-corruption
By André Petheram and Isak Nti Asare
In the fight against corruption, authorities need every tool they can find. It is an expensive crime, with an estimated US$1.5 to $2 trillion lost each year to bribes globally. The problem is consistently difficult to solve, with most countries showing little improvement in Transparency International’s latest Corruption Perceptions Index.
However, thanks to the sustained efforts of civil society, governments, and businesses as part of the open data movement across the world, it is getting easier to see how governments spend their money. Open information about government contracts with private companies, a common source of corruption, is now consistently available in many countries. Ukraine’s innovative ProZorro portal, for example, helps anti-corruption bodies spot tenders designed to favour particular bidders.
Governments themselves are playing an important role in working out which data is needed to combat corruption, with Mexico’s implementation of the Open Data Charter’s Anticorruption Open Up Guide leading the way. Moreover, as more of this data is opened, data mining has an increasingly important role. This underpins research projects such as the pan-European Digiwhist, which analyses millions of EU government contracts to identify those particularly vulnerable to abuse.
The natural next step is to explore the use of artificial intelligence and machine learning tools to find insights and anomalies within big and open datasets. We think AI will have three main benefits in fighting corruption.
First, artificial intelligence can reveal patterns too complex for humans to see without mechanical assistance. In 2017, for example, Spanish researchers Félix López-Iturriaga and Iván Pastor Sanz used neural networks to build a predictive model for corruption in Spanish provinces. This algorithm identified a previously unseen relationship between particular economic factors, such as rising real estate prices, and corruption cases. The model can identify corruption before it emerges, potentially allowing authorities to take pre-emptive measures. The next challenge will be to train algorithms on yet more complex and diverse datasets in a search for deeply hidden corruption indicators. Moreover, researchers and developers must begin co-designing their tools with authorities, so that these innovations can be put to immediate use in everyday anti-corruption work.
Secondly, AI’s ability to process large amounts of data allows people to focus on details. Anti-corruption analysis is labour-intensive. As more and more data becomes available to anti-corruption organisations, it becomes harder to work through it all. Artificial intelligence programmes can act like inexpensive analysts for teams that are often poorly-funded, and can work in real-time. In the related field of fraud detection, for example, Citibank has invested in a machine learning platform that flags suspect transactions as they occur. If artificial intelligence can be used to enhance, deepen and accelerate routine data analysis, then people can be freed to scrutinise suspicious contracts or payments in depth. This can increase the rate of corruption prosecutions.
Finally, you can’t threaten an algorithm. Anti-corruption activists’ bravery and endeavour has often been the most important factor in holding governments to account and forcing the prosecution of corrupt individuals. However, if authorities and activists increasingly use artificial intelligence, this will mean that less anti-corruption activity is vulnerable to violence and legal pressure. Algorithms can largely churn away no matter what is going on externally. This effect could be intensified by ensuring that any anti-corruption AI tool is overseen by people far away from the targets of surveillance.
Governments, NGOs and law enforcement agencies are not AI ready when it comes to anti-corruption. The number of researchers working in the field is small. Most crucially, developers and anti-corruption bodies must develop a rigorous understanding of which datasets can most help AI practitioners: for example, is it more useful to look at government spending, or to access the financial accounts of suspected individuals? Oxford Insights will begin research into this question in the coming weeks, and we would welcome any thoughts and suggestions that are out there.
Artificial intelligence has the potential to recapture millions, and possibly billions, of dollars for governments and therefore citizens across the world. However, the work required is yet to begin in earnest.