The nature of technologies, as a non-neutral construct, implies the incorporation of the cultures and values of those who design and develop them, from which artificial intelligence technologies are not exempt, including the Silicon Valley culture.
During a year-long research and coordination project, nineteen data research experts from across Latin America and the Caribbean compiled what can be called the first comprehensive inventory of 'AI in Latin America and the Caribbean'. The volume provides a wealth of knowledge, analysis, and policy recommendations, and is helping to kick-start a research field that focuses on and transcends its regional context. GuIA is a project of the Centro de Estudios en Tecnología y Sociedad (CETyS), Universidad de San Andrés, Argentina.
Dr Carolina Aguerre, who initiated and acted as editor of the volume, is a co-director of CETyS and is continuing on this path in her current research as a Senior Research Fellow at the Centre. Speaking about the origins of the project, Aguerre explains that there is a need to provide visibility to the conceptual underpinnings and practical challenges emerging from a region with different characteristics from that of developed economies where most AI systems are developed. This academic production would feed into the AI Observatory (fAIr LAC) at the IADB. Only this kind of coordinated interdisciplinary research could successfully deal with an issue that spans different levels of expertise. At the same time it is more than a purely academic endeavour. Traditional modes of governance are challenged in a regional context. Aguerre reflects about 'modular governance' as a technique to manage complex systems. She engages with theorists Gasser and Almeida, who propose three interconnected levels for data, ethics and regulation. However, the goal is to rethink this model and develop it further in the context of AI.
Countries vary widely. One of the fascinating aspects in reading the compilation volume is the variety of ways that particular countries are responding to the AI innovation on the horizon. Health and social classification systems are at the core of the challenge. In Columbia, SISBEN, a social classification system, is being discussed. Here, the 'poor' person is created by an algorithm and defined as a subject of social benefits (López and Castañeda). It is agreed that regulatory sandboxes are a feasible way to (de-)regulate innovation in defined experimental frames. One contribution proposes a regulatory sandbox in an innovative way to test the ‘explanability’ of AI, because 'privacy, agency, autonomy and self-determination are essential rights sitting at a crossroad between ethical principles and legal rules' (Castaño).
When looking at individual countries, it is clear that government institutions, civil society, and academia are involved to differing degrees—a disparity that can produce tensions. Aguerre is convinced that though civil society organizations publish high impact documents, they are sometimes 'empirically not well-founded'. She cherishes the academic freedom of a well-funded project, which combines motivations from institutional donors with those of academics. Regional and global perspectives are intertwined here.
Aguerre, who has scrutinized national AI strategies and data governance in her own contribution to the volume, reflects about harmonization of legislation or standards. Would it be desirable among countries in a regional context? In her view, national policies prevail. 'It is a construction to talk about Latin America in terms of policy.' What is at stake? Strong asymmetries in data governance policy may intensify; they open up anew between data generators and solution providers, reflecting similar asymmetries on the global scale.
Is there a global platform for deliberation and coordination on the governance of AI? The initial steps of the Global Partnership of AI (GPAI), a work conducted with the support of the OECD, may provide a future blueprint for a multidisciplinary process with government oversight to agree on guiding rules for responsible applications of AI technology. The OECD is working with the French, German and Canadian governments among twelve other States on the GPAI and Aguerre is participating in one of the working groups on 'Responsible AI'. Aguerre sees the OECD expanding its agenda on AI governance with a greater dynamism than many UN bodies. From her view
'The leadership of this goes well beyond the membership of the OECD. It will probably become a more efficient organization in transposing those principles than many UN organizations. Industry also sees strength in promoting the work of the OECD on AI governance. There is a lot of sponsorship from the private sector, seeing the OECD as an efficient secretariat to promote this. Yet, the challenge of inclusivity and representativeness of non-OECD countries should be thoroughly acknowledged if GPAI wants to have an impact on global AI governance’.
Aguerre also this year participated as a member of UNESCO's ad hoc expert working group to draft a document on AI recommendations for the application of ethical principles. The document is now being reviewed by governments before the final approval in August 2021. This document was recognized as one of this year's most impactful AI ethics initiatives.
Artificial Intelligence in Latin America and the Caribbean. Ethics, Governance and Policies.
ISSN 2684-0278 Buenos Aires, Argentina.
Editor: Carolina Aguerre
Authors: Carolina Aguerre - Carlos Amunátegui Perelló - Chelcée Brathwaite - Juan Diego
Castañeda - Daniel Castaño - Claudia Del Pozo - Lorena Flórez Rojas - Constanza Gómez
Montt - Juan Carlos Lara Gálvez - Joao López - Raúl Madrid - Ana Victoria Martín del Campo -
Juliana Vargas Leal
Gasser, Urs, and Virgilio A.F. Almeida (2017). 'A Layered Model forAI Governance.' IEEE Internet Computing 21 (6) (November): 58–62. Doi:10.1109/mic.2017.4180835