Algorithmic transparency of automated decision-making systems: A case study of the Colombian government

Only a handful of national and subnational governments around the world actively inform citizens about their use of automated decision-making systems (ADS). For example, few national and subnational governments have made available public registries of the algorithms they use. Other governments disclose the use of specific algorithms that support their decision-making processes but limit the availability of information about how the algorithms operate. Furthermore, although the adoption of artificial intelligence (AI) and other algorithmic technologies in the public sector is a global trend, most of the literature on the implementation of ADS by governments focuses on Global North countries. This chapter aims at contributing to bridge this gap by analyzing the levels of transparency with regards to the acquisition, development, piloting, adoption, and use of ADS in Colombia’s public sector. On one hand, this chapter assesses the Colombian governments’ repositories of public algorithms and, in general, the disclosure of data on the use of AI and robotic process automation (RPA). On the other hand, the chapter presents the results of a novel database of 113 ADS adopted by the Colombian public sector that allows to assess the availability of information about these systems. This research follows a case study approach in which quantitative and qualitative data about algorithmic transparency in Colombia was collected and processed. The primary data was drawn from pre-existing public databases and repositories and primary information published by public entities in their annual management reports, institutional web pages, social networks, and press reports. Secondary sources included information published by tech companies, news reports, grey literature –published by multilateral entities and civil society organizations– and academic literature. We report that national government of Colombia has created three online AI repositories which include detailed information, but there is a significant under-registry of the systems adopted by the public sector. However, we also found that the basic features of most of the systems included in our novel database can be documented through information made public by government organizations through other means. Despite of using all publicly available primary and secondary sources to characterize each ADS adopted by the Colombian public sector, we report that there is very little information on critical aspects such as the type of data used by the systems, the performance of the algorithms, their cost, whether the software is proprietary or open-source, and the procurement processes required to acquire the respective license and/or service. The information that is publicly available about the systems is insufficient to make the algorithmic processes truly accountable in Colombia.