Tensions in the AI race don’t necessarily foreshadow doom, but they are the consequence of a game of imperfect information. Jeanne Marie Jacqueline Vincendeau explains that the framework of any game based on Bayesian theory is neutral. The mistrust between China and the US arises from the misinterpretation of each other’s behaviour
The three features of Bayesian games
A key theme in the AI innovation race is uncertainty. Governments consider each other black boxes. Their knowledge about how and when to act, and what impact those actions might have, is limited. The players do not know the entire game tree, nor exactly which game they are playing. Governments take decisions quasi-blindly, maximising self-interest in the blurry landscape of frontier AI.
The one known variable is the players’ nature – and this guides strategy. Their nature is formed from an entanglement of historical pathways to collectivism and individualism, and national ideology. AI innovation in China focuses on centralised cooperation and decentralised competition with the US. This guides expectations, and sets odds for each player’s moves.
A key theme in the AI innovation race between the US and China is uncertainty, but the one known variable is the players’ nature – and this guides strategy
Bayes’ theory is based on the ability to update probabilities. New information strengthens previous knowledge or assumptions. Over time, updating beliefs allows us to develop more precise expectations. China and the US follow this pattern of recursive adaptation and updating. Their beliefs about each other are based on the history of previous ‘moves’. The two superpowers modify their actions to keep up with new information. Rapid breakthroughs and potential threats guide the gamers’ positionality.
Coping with uncertainty starts at the origin of the problem
A Bayesian game is ‘an extensive-form game of imperfect information‘. Its outlook on human nature is less pessimistic than the prisoner’s dilemma – a thought experiment involving two rational agents, each of whom can either cooperate for mutual benefit or betray their partner for individual gain.
The Bayesian game is not skewed towards collaborative or competitive behaviour. The payoffs adapt and evolve based on the players’ dynamics. In relations between China and the US, a lack of information fuelled mistrust, and rapid innovations triggered new anxieties, such as concerns about domestic performance or the competitor’s use of AI. Together, they tarnished Sino-US geopolitics. The evolution of players’ expectations and intentions slowly reveal themselves throughout the iterations of the game over time. Mutual distrust or mistrust is not a symptom; it has been (more or less) involuntarily crafted.
This modelling of the AI governance struggle between Beijing and Washington enable us to draw more accurately the visions and expectations of Chairman Xi and President Trump, making it easier to study and simulate the next steps.
In a game of imperfect information, modelling the AI governance struggle between China and the US paves the way for forecasting, which may help lower uncertainty
The modelling offers us insights to better understand decision-making, and paves the way for forecasting. It may lower uncertainty and make the odds evolve in a more positive way.
Player, prepare your next move
The faster Sino-US relations improve, and tensions deescalate, the safer the global system in relation to AI geopolitics. To achieve this, players must adapt to different roles.
The primary uncertainties about China are its political stance and its application of AI in military operations. In 2022, Beijing claimed it was rejecting the possibility of using AI Lethal Autonomous Weapons Systems. Nevertheless, it has recently been seeking to implement ‘AI-enabled battlefield command and control‘ to provide more autonomy to its weapon systems. Thus, Beijing’s leadership should update its official position to enhance mutual understanding.
To achieve safer AI geopolitics, China must address its application of AI in military operations. In the US, the uses and abuses of AI tools in transborder conflicts and domestic surveillance must end
In the US, the government lacks accountability, domestically and internationally. The uses and abuses of AI tools in transborder conflicts and domestic surveillance are already affecting lives. Legal frameworks with checks and balances could reestablish trust in Washington’s power.
A secondary concern about both leading AI players is their grim trigger strategy for a repeated game. Avoiding a point of no return, and offering a clear definition of which limits neither side may cross, requires a bilateral agreement.
Thirdly, the international community – a multitude of smaller players with varied and sometimes unknowable natures – will become ever more involved in Sino-US tensions. This community’s mediatory role will be fundamental to navigate rapidly escalating tensions.
Middle-sized powers can offer neutral fields or infrastructure, non-ideologically aligned advice, or support in bridging AI safety gaps. These powers need not be governments (such as those of the EU and India in particular), but academia and civil society organisations, too.
Global AI governance: the endgame
In 2025, China proposed creating the first global AI cooperation organisation. Given the geopolitical tensions in the AI race, it would be surprising if China’s project achieved the global reach it intends.
The United Nations could be an alternative platform to global AI governance. But veto power and competing ideologies have, historically, hindered the UN’s decision-making process. AI governance must be forward-looking and adapt to rapid innovation.
In such an uncertain game, the only means to avert a downward spiral are good faith and proactivity. China, the US, and other capable players can all help write the rulebook to improve trust.
Agreeing on, as well as ratifying, clear rules will provide more certainty and decrease mistrust. Only then will assumptions of the players become more secure and positive, allowing a deescalation of current tensions.