During the IDB’s 2023 Knowledge Week, economists Michael Kremer and Esther Duflo highlighted the importance of working in tandem with academics and promoting innovative initiatives, for producing robust evidence regarding potential solutions to the different problems which are currently affecting our region.
Among these problems, bureaucratic corruption is a leading one; according to estimates by the IMF, corruption costs countries about 4% of their GDP worldwide. This phenomenon also adversely impacts the quality of public and foreign direct investment, as well as countries’ credit ratings, among other things, further increasing the costs of corruption.
In addition to the aforementioned economic costs, corruption scandals are sources of political instability and contribute to a reduction in public trust in institutions, as well as in the democratic foundations of our government systems.
In spite of its importance, evidence of effective solutions in terms of corruption control is still at its early stages. Because of the very nature of corruption, governmental reforms within the spirit of promoting transparency and eliminating this phenomenon lack a unique and solid dimension, which limits the possibility of designing and implementing causal frameworks to evaluate their respective effects.
This is the very reason why the joint effort of academics and innovation centers, focused in generating real impact by applying the former’s cutting-edge research, becomes a central pillar of the fight against corruption.
An innovative and collaborative strategy to produce evidence on corruption
A particular example of such efforts is that of the Peruvian government’s Comptroller General’s Office (CGR). By leveraging new technologies, this corruption oversight agency is implementing, together with the technical and financial support of the IDB, a loan-program (PE-L1240) destined to improve the effectiveness of its government corruption control services.
This joint effort has already produced preliminary evidence on the effectiveness of public awareness tools, similar to those of MapaInversiones in terms of Citizen Monitoring Controls, in reducing overbilling in third-party public infrastructure projects, as well as those of New Talents in Government Control, in terms of the efficacy and analysis of whistle-blower allegations of corruption.
Currently, the IDB has struck an agreement with Columbia University to collaborate on the design of experimental strategies, the most robust way of proving causality, in order to inform the modernization of the CGR’s internal system for receiving and analyzing citizens’ whistle-blower allegations, incorporating models of artificial intelligence and machine learning.
The interventions and experimental studies stemming from this collaboration will be monitored by the Laboratory of Evidence in the Control, financed also by this project, which is currently known as the National Anti-Corruption Observatory. This observatory would incorporate said practices as tools to improve its capacity in generating evidence for other areas lying within the program, replicating research techniques used by IDB specialists and Columbia University. The National Science Foundation has rewarded this initiative for its innovative design.
Prioritization and admission algorithm: Opportunities for strategic use of resources in detecting corruption
The main issue which this experiment seeks to address is how to raise the probability of correctly identifying corruption from the more than 5,000 whistle-blower allegations made by citizens, through different channels, to government agencies.
The sheer scale of this information flow motivated the idea of developing algorithms to facilitate the processing and analysis of such allegations. The first algorithm, known as the “admission algorithm”, identifies and filters out allegations which are not under the jurisdiction of the CGR.
According to estimates for the year 2022, approximately 40% of the received allegations do not necessitate CGR consideration. Therefore, this algorithm evaluates whether a given allegation meets the three required criteria for admission, retaining those which do. These considerations include identifying the person who is accused of a corrupt dealing, and providing a description of said corrupt dealing, the affected party, and the date of occurrence.
The second algorithm, called “priority algorithm”, seeks to improve the classification, and the consequent hierarchical ordering, of the allegations. Using probability, time, and incident amount algorithms, it evaluates the probability that the allegations actually involve corrupt dealings, as well as the monetary amounts under dispute. These various components result in an associated score for each allegation, where higher scores denote greater priority for further analysis. Such scores provide analysts and managers with statistical justification for the choice of which allegations to pursue.
The use of these machine learning algorithms and artificial intelligence in such ways promises to radically transform the way in which corruption allegations are classified and analyzed by the CGR, while at the same time, it will allow us to draw valuable conclusions on how government control agencies in other countries in the region may strengthen their strategies for preventing and fighting corruption.
These developments showcase the growing reach and importance of new technologies in the design and development of public policy, as well as in the implementation of government control mechanisms, which allows for more informed and efficient decision making in the public realm. The implementation of these tools has been designed since the beginning of the project to unfold in a randomized manner, which will allow to generate experimental causal evidence on its impact. This is possible thanks to the work of an interdisciplinary team of specialists from the IFD/ICS and SPD economists from the IDB.
Such a strategy of producing impact from the experimental design, with clear goals and objectives, as well as through the use of innovative technologies focused on the generation of research evidence, and in collaboration with prestigious universities, has been replicated in other contexts and countries, achieving successful results. In the next blog, we will discuss the ongoing innovations and impacts on Digital Justice in Brazil, driven by the IDB in collaboration with Harvard University.