AI in the Crosshairs: How Geopolitics Shapes Technology
Few technological advances have captured the public imagination like the recent boom in generative artificial intelligence (AI) power and the Large Language Models (LLMs) that power them. Although this may seem like another tech cycle we believe this technology is unique due to its inherent decentralized governance and the increased geopolitical focus it has already attracted in its early adoption phase. In our view, recognizing how these distinctions will influence the technology's evolution is crucial for understanding its potential implications for business and global relations.
Artificial Intelligence is viewed by some political leaders as a technology that potentially carries greater geopolitical significance than previous technological shifts. Earlier transitions, like the adoption of cloud computing, expansion of social media and the proliferation of smartphones, rose initially with minimal governmental interference. Contrarily, the strategic status of AI stems from its "dual-use" nature, meaning the same technology has both civilian and military applications. In a recent speech mentioning AI, White House National Security Advisor Jake Sullivan stated "Preserving our edge in science and technology is not a 'domestic issue' or 'national security' issue. It's both.” 1 The intelligence community and US Department of Defense are leveraging advancements in this field for real-time surveillance and threat detection, and AI stands to play a central role as increasingly autonomous systems transform defense technology.2 Additionally, we believe LLMs in particular could potentially enable a new generation of cyber weapons, enhancing existing cybersecurity threats like phishing attacks, system hacks, and manipulated media. This trend is highlighted by the recent success of teams that use LLMs in cybersecurity competitions.3
Given the strategic applications of AI, this technology is now a key factor in the ongoing competition between the United States and China. The US has taken action to restrict China’s access to advanced AI by imposing export controls specifically on the high-performance Graphics Processing Unit (GPU) chips that are required today to develop cutting-edge AI systems.4 China responded back with national security related export restrictions on gallium and germanium which are used to produce chips, solar panels, and fiber optics.5 With AI leadership a clear priority for both the US and China, critical commodities for AI may continue to be subject to export restrictions. AI will likely receive similar treatment to other dual-use technologies like nuclear and aerospace.6 From the emergence of an escalating high-tech trade competition, to government subsidization driving the boom in Chinese semiconductor startups and American chip manufacturing, state involvement has already begun to reshape global markets.7
In addition to US export controls, we believe the information environment in China could hinder their progress in LLMs. LLMs depend on vast volumes of text for development, and constraints on permissible content could compromise the performance of Chinese LLMs compared to western LLMs that have ready access to the full open web. Furthermore, outputs of LLMs are unpredictable and difficult to control. The Chinese government’s focus is evident in a recent regulatory framework announced by the Cyberspace Administration of China, which aims to impose disclosure requirements, liability for output, and restrictions on data used in the development of LLMs.8 We believe these policies underline the central challenge all countries face with LLMs: balancing the pursuit of strategic technology leadership with controlled deployment to maintain domestic stability.
In our view, national actors cannot control all the potential futures for AI. Most of today’s prominent LLMs are “closed” systems owned by private companies like OpenAI and Google. However, we are seeing the simultaneous rise of open-source LLMs which may be used and modified permissionlessly by anyone with internet access. This rise is, in part, the result of major corporations such as Meta committing to open-source LLMs as part of their AI strategy.9 The potential benefits of open LLMs are significant: they could create a more competitive, accessible, and consumer-friendly market for AI while also improving auditability of LLMs and their outputs. Simultaneously they introduce new risks, most obviously the potential for misuse of uncensored models which can be easily created by individuals.10 These open-source LLMs are improving rapidly; with over forty of them performing comparably to ChatGPT in specific benchmarks.11 We believe if these trends continue, it could drive further commoditization of “closed” LLMs, weakening the competitive advantage of large AI companies, while also dampening the impact of US export controls on GPUs that were designed to limit China’s access to advanced LLMs. Regardless, even if open-source LLM gains plateau, the current level of progress suggests that a baseline level of LLM capability will soon be accessible to all, irrespective of restrictions.
The swift pace of change raises three key questions at the intersection of geopolitics and LLMs.
- How will other countries align in the geopolitical competition surrounding AI? The Netherlands and Japan following the US lead on chip exports serve as notable examples.
- How do we balance access to innovation versus safety? Said another way, how can we prevent dangerous capabilities from falling into the wrong hands, while ensuring AI doesn't end up monopolized by a few large corporations?
- How might principles of nonproliferation, traditionally applied to physical entities like nuclear materials, evolve with digital-native technologies like AI?
We believe the rapid adoption of LLMs suggests these questions may confront us sooner rather than later. The way governments and citizens respond to these challenges, and the unforeseen twists in AI development, may shape geopolitical and market dynamics for years to come.
1 Remarks by National Security Advisor Jake Sullivan at the Special Competitive Studies Project Global Emerging Technologies Summit. As of September 16, 2022.
2 National Security Commission on Artificial Intelligence Final Report. As of 2021.
3 Wall Street Journal, “ChatGPT Helped Win a Hackathon.” As of March 20, 2023.
4 United States White House, “Fact Sheet: CHIPS and Science Act Will Lower Costs, Create Jobs, Strengthen Supply Chains, and Counter China.” As of August 9, 2022.
5 Foreign Policy “China Fires a Fresh Salvo in the Chip War -Foreign Policy.” As of July 6, 2023
6 Center for Strategic and International Studies, “Choking off China’s Access to the Future of AI.” As of October 11, 2022.
7 Center for Strategic and International Studies, “Choking off China’s Access to the Future of AI.” As of October 11, 2022.
8 Cyberspace Administration of China: Measures for the Management of Generative AI Services.” As of April 11, 2023.
9 Meta AI, Llama 2 Announcement. As of July 18, 2023.
10 New York Times, “Uncensored Chatbots Provoke a Fracas Over Free Speech.” As of July 2, 2023.
11 Hugging Face, Open LLM Leaderboard, Edward Beeching, Clémentine Fourrier, Nathan Habib, Sheon Han, Nathan Lambert, Nazneen Rajani, Omar Sanseviero, Lewis Tunstall, Thomas Wolf. As of 2023.