Can apparent bystanders distinctively shape an outcome? Global south countries and global catastrophic risk-focused governance of artificial intelligence (2312.04616v1)

Abstract: Increasingly, there is well-grounded concern that through perpetual scaling-up of computation power and data, current deep learning techniques will create highly capable artificial intelligence that could pursue goals in a manner that is not aligned with human values. In turn, such AI could have the potential of leading to a scenario in which there is serious global-scale damage to human wellbeing. Against this backdrop, a number of researchers and public policy professionals have been developing ideas about how to govern AI in a manner that reduces the chances that it could lead to a global catastrophe. The jurisdictional focus of a vast majority of their assessments so far has been the United States, China, and Europe. That preference seems to reveal an assumption underlying most of the work in this field: That global south countries can only have a marginal role in attempts to govern AI development from a global catastrophic risk -focused perspective. Our paper sets out to undermine this assumption. We argue that global south countries like India and Singapore (and specific coalitions) could in fact be fairly consequential in the global catastrophic risk-focused governance of AI. We support our position using 4 key claims. 3 are constructed out of the current ways in which advanced foundational AI models are built and used while one is constructed on the strategic roles that global south countries and coalitions have historically played in the design and use of multilateral rules and institutions. As each claim is elaborated, we also suggest some ways through which global south countries can play a positive role in designing, strengthening and operationalizing global catastrophic risk-focused AI governance.

Analysis of Global South Influence on AI Governance and Global Catastrophic Risk

The paper "Can Apparent Bystanders Distinctively Shape an Outcome? Global South Countries and Global Catastrophic Risk-Focused Governance of Artificial Intelligence" by Cecil Abungu, Michelle Malonza, and SumayaNur Adan presents a compelling analysis of the underestimated role that global south countries could play in AI governance with respect to global catastrophic risks. This topic, while often dominated by the geopolitical dynamics of the United States, China, and Europe, is scrutinized with a fresh perspective that highlights the strategic significance of countries in the global south.

Core Contributions

The authors construct their argument around four pivotal claims that advocate a consequential role for global south countries in the governance of AI:

1. Technical Capability and Infrastructure: The paper notes the potential of global south countries, like India and Singapore, to contribute significantly to AI development via advanced computational infrastructure. India's investment in supercomputing, alongside its involvement with major tech companies like Nvidia, positions it as a key player in the global AI landscape.
2. Human Feedback in AI Training: The authors emphasize the integral role of Reinforcement Learning from Human Feedback (RLHF) in training highly capable AI systems. The prevalence of data labeling operations in global south countries, driven by cost efficiencies, underscores their critical involvement in shaping AI outputs.
3. Open-Source Development and Fine-Tuning: The paper argues that global south countries have the requisite talent and technical expertise to fine-tune frontier AI models. Resources like Low-Rank Adaptation (LoRA) and the development of advanced models echo the region's capabilities in affecting AI trajectories.
4. Strategic Multilateral Influence: Historical precedents show global south countries effectively shaping multilateral institutions and rules, as seen in climate negotiations and intellectual property discussions. This suggests potential influence over global AI governance narratives.

Implications and Future Prospects

The research presented harbors both theoretical and practical implications. On a theoretical level, it challenges conventional notions about the loci of influence in AI governance, suggesting that the focus should shift to include these geographically and politically diverse regions. From a practical standpoint, it lays the groundwork for developing regulatory frameworks and policies that engage global south countries as active stakeholders in preventing AI-induced global risks.

The paper proposes specific regulatory pathways for these nations to mitigate potential risks associated with the development of highly capable AI models. These include implementing stringent reporting requirements for large-scale AI training and bolstering labor rights for data labeling personnel to increase the cost and hence caution the development of potentially harmful AI capabilities.

Speculative Outlook

Looking forward, a continued investigation into how global south countries can adopt and influence AI governance mechanisms is warranted. These regions could potentially spearhead unique regulatory frameworks that might serve as a blueprint for global initiatives. Encouraging a nuanced dialogue about AI's role in economic development versus its existential risks could transform global policy approaches, driving more inclusive and sustainable AI advancements.

Conclusively, the authors provide a substantive argument for expanding the scope of AI governance discussions to include global south narratives, leveraging their technical capabilities, influential roles in rule-making, and contributions to AI's evolving landscape. This shift in focus is not just a matter of ethical consideration but also a strategic necessity in anticipating and mitigating the complex risks posed by advanced AI.