The Geopolitics of AI | Global Conference 2024

Geopolitical Framing and Strategic Competition

• The U.S.-China AI relationship is characterized as a strategic competition by both U.S. Democrats and Republicans, though China officially rejects this term in favor of "win-win cooperation."
• Chinese leadership, specifically under Xi Jinping (circa 2017-2018), identified AI as a "game-changer" for global economic competitiveness and military power, triggering a concentrated industrial policy effort.
• Despite China's focus on dominance, there is internal anxiety within the Chinese Communist Party regarding the use of generative AI for domestic political destabilization (e.g., queries on removing leadership).

Assessment of the "Race" and Current Standing

• Multiple parallel races are occurring: building the largest powerful models (led by the U.S. private sector), developing small purpose-built models (e.g., for agriculture), and creating defensive cyber capabilities.
• The U.S. and its allies (UK, EU) are currently ahead in building the most powerful foundation models, though China is rapidly closing the gap in specific hardware areas.
• China's proposal for high-level U.S.-China dialogue in November 2023 is interpreted as a signal of concern regarding their relative standing in the ecosystem.
• Karen Pierce notes that while the U.S. and allies may have initial setbacks (similar to the Soviet "Sputnik" moment), long-term competitiveness relies on open societies and alliances, though the timeline for pulling ahead remains uncertain.

Export Controls and Supply Chain Security

• The U.S. maintains a "small yard, high fence" export control regime to restrict China's access to advanced semiconductors, a strategy begun under the Trump administration and expanded by the Biden Executive Order.
• The "foreign direct product rule" allows U.S. export controls to reach foreign-made products if they utilize protected U.S. technology inputs.
• Allies Japan and the Netherlands have joined the export control regime to deny China access to key enabling technologies like lithography machines.
• The appearance of Huawei's 7nm chips suggests the U.S. strategy is a "slowdown operation" rather than a complete prevention of Chinese advancement.
• There is ongoing debate regarding the relevance of WTO rules, with officials noting China has historically bypassed multilateral trade norms through IP theft and forced technology transfer.

Regulatory Divergence and the EU AI Act

• The EU AI Act represents the first comprehensive regulatory standard, positioning the EU to potentially set global norms through the "Brussels Effect" given the EU's market size.
• The EU approach prioritizes human rights, safety, and democratic values, accepting that strict regulation may stifle innovation compared to less regulated jurisdictions.
• Kevin Rudd advocates for a "Swiss posture," seeking a compromise between the EU's privacy-focused regime and the U.S. innovation-focused approach to effectively harness data.
• Policymakers express concern that the EU currently acts more as a "referee" than a "player" in fostering competitive tech ecosystems.

Governance, Private Sector Role, and Multilateralism

• A key distinction from the nuclear era is that AI innovation is driven primarily by private companies rather than governments, requiring new public-private partnership models for governance.
• The U.S. administration employs a three-part strategy: voluntary safety commitments from tech companies, an Executive Order outlining risk assessments and promotion of use cases, and potential new legislation.
• There is a push to increase "fluency" between the public and private sectors by staffing government agencies with technology experts and creating shared understanding of regulatory frameworks.
• Plurilateralism vs. Multilateralism: Panelists favor plurilateral approaches (like the OECD or alliance of democracies) over UN-wide multilateralism, which they fear would empower China, Russia, and Iran.
• Proposed "variable geometry" negotiations focus on narrow areas of mutual interest, such as banning the removal of human agency in lethal autonomous weapons systems.
• Some panelists suggest creating a new international body similar to the International Energy Agency (IEA) to manage critical minerals, chips, and compute power against Chinese influence.

Risks: Disinformation, Elections, and Weaponization

• AI is already weaponized via social media algorithms that enable high-volume, tailored messaging and propaganda, raising concerns about upcoming elections in the U.S., India, UK, Canada, and Europe.
• Specific risks include the use of AI for deepfakes, non-consensual intimate imagery (affecting female political leaders), and automated spear-phishing campaigns.
• While a single "deepfake event" shaking a democracy has not yet occurred, the cumulative impact of AI-generated disinformation is described as a pervasive, minute-by-minute threat.
• Countermeasures include the "Tech Accord" among companies to watermark content and use AI to detect deepfakes, though panelists note coordination on misinformation remains insufficient compared to cybersecurity protocols.
• Kevin Rudd warns that democracies must move from passive observation to actively imposing costs on hostile foreign actors engaging in subversion.

Data and Privacy Considerations

• The assumption that China has a massive data advantage due to population size is contested; while China has large volumes of data (e.g., in oncology), the quality and usability for global language models are debated.
• Western democracies prioritize individual privacy and civil liberties in data handling, potentially creating a disadvantage in raw model training speed compared to authoritarian regimes, but a long-term advantage in public trust and adoption.