The Geopolitics of AI: Large Language Models
What are the geopolitical implications of nations competing for advanced large language models?
Large Language Models (LLMs) are the latest iteration of AI on our ever-expanding technological frontier. What makes them so valuable are the various geopolitical applications they have across domains of strategic importance.
The dual-use nature of LLMs for malicious or benevolent purposes and ability to execute on those tasks with incredible speed is what makes them formidable - and critical for military and economic security.
Here is a brief primer on these AI systems, written - rather appropriately - by the system itself:
“Large Language Models (LLMs) like GPT-4 work by analyzing and generating text based on patterns they've learned from a vast database of written language. Imagine them as highly sophisticated pattern-recognition systems. They're not conscious, nor do they understand language in the human sense, but they can mimic understanding by drawing on a huge array of examples they've been trained on”.
The commercial use cases of LLMs across industries and sectors are so deep and wide that an entire book could be written on them. Instead, the focus will be on their applications in a broad geopolitical context.
Military and Security
Defensive
During geopolitical crises, LLMs can aid in real-time information gathering and analysis, providing critical insights for decision-makers. They can also assist in managing the flow of information to the public and international community.
LLMs can process and analyze vast amounts of data, including news, social media, and other open-source intelligence, to identify trends, patterns, and insights that could inform tactical strategies in real time.
They can be also used to enhance cybersecurity protocols by identifying and responding to cyber threats. This would involve integrating LLMs into an organization’s cybersecurity system and leveraging the existing pool of data within a specific context to generate the optimal defense protocols for a particular agency.
During geopolitical crises, LLMs can aid in real-time information gathering and analysis as well as managing the flow of information to the public and international community. However, all of these use cases also have offensive applications.
Offensive
In the context of cyberwarfare, LLMs could also potentially be used in the development of sophisticated cyber-attacks or defense mechanisms.
Hostile foreign powers can use LLMs to analyze and undermine public opinion on a global scale. They could be employed to create persuasive content or propaganda tailored to specific audiences in different geopolitical contexts with far more speed and persuasion than current methods.
Beyond traditional propaganda, LLMs could also be employed to create and spread disinformation at scale. These are familiar risks: fake news articles, social media posts, or other forms of content designed to mislead, confuse, or manipulate public perception or decision-making processes.
Countries with a legacy and history of democratic institutions like the US or UK could see these processes disrupted.
LLMs could be used to create and manage a large number of social media accounts (often referred to as "bots") to amplify certain viewpoints or narratives, effectively swaying public opinion or political discourse in a manner beneficial to specific military or geopolitical goals.
And this is only scratching the surface.
Given the wide-ranging applications of LLMs, it is therefore no wonder that AI and semiconductors are at the heart of the US-China competition. Particularly for Beijing, which relies on so-called “asymmetric warfare” and a model of military strategy known as “Systems Confrontation”.
All of these require vast amounts of data, and software systems powerful enough to leverage them optimally.
China
China faces hurdles in indigenous innovation in LLMs, partly due to its tech crackdown. Issues like censorship and content liability pose significant challenges, potentially hindering progress and innovation.
Furthermore, U.S. restrictions on semiconductor exports impact Chinese LLMs, which rely heavily on advanced chips for computing power; one such example are Nvidia’s A100 chips.
Europe
Europe trails in the AI race, lacking key success factors evident in the US and China. The absence of a robust VC-entrepreneur ecosystem, successful consumer internet and social media companies, and substantial government regulations in AI hampers Europe's competitiveness in the field.
For a more detailed breakdown on AI development, governance, and obstacles between the US, China, and Europe, see my previous reports:
United States
The US continues to boast the most developed LLM systems and highest volume of capital invested in the technology. The investment in the U.S. was approximately $47.4 billion, which was about 3.5 times the $13.4 billion invested in China.
In the realm of AI startups, the U.S. has taken a clear lead, boasting nearly twice as many newly funded companies as the combined total of the European Union and the UK, and outnumbering those in China by 3.4 times.
American dominance in GPU design and production, a critical component for LLMs, further solidifies their competitive edge.
But Washington cannot rest on its laurels.
Pay Attention
Unlike nuclear technology, the decentralized nature of AI, characterized by open-source algorithms, democratizes access. India, China, and the European Union are actively using state support to advance their technological capabilities, particularly in the field of AI.
Although they currently trail behind the United States in this domain, these regions are focusing on enhancing their LLM development to minimize reliance on external technologies and to secure a competitive edge in this rapidly evolving sector.
A report by Goldman Sachs concluded that:
We are also 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.
These open-source LLMs are improving rapidly; with over forty of them performing comparably to ChatGPT in specific benchmarks. 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.
Consequently, the democratization of LLMs could erode the competitive edge Washington currently enjoys, and with it, a degree geopolitical leverage. In an increasingly multipolar world, every bit and byte counts.