Amid rising tensions over Taiwan, a looming nuclear arms race and tit-for-tat espionage allegations, the US and China have agreed to crucial talks on the future use of artificial intelligence (AI) in sensitive military areas.
US President Joe Biden and Chinese President Xi Jinping affirmed the need for bilateral government talks to address the risks associated with using advanced AI systems in sensitive military functions at this week’s Asia Pacific Economic Cooperation (APEC) summit in San Francisco. But it was not clear from post-meeting statements that the two sides are anywhere near making a joint declaration.
The South China Morning Post (SCMP) reported citing anonymous sources that Biden and Xi were set to pledge a ban on using AI in nuclear weapons command and control (C2) and autonomous weapons such as drones. The report said that the US and China are concerned about the potential for unregulated AI to fuel and escalate conflicts.
Oriana Skylar Mastro, a fellow at Stanford University’s Freeman Spogli Institute for International Studies, said in the SCMP report that one of the first agreements the two sides should strike is to avoid automating nuclear C2 systems with AI.
Mastro pointed out that humans currently handle decision-making and launch procedures, although she said there are discussions underway about having AI automate parts of these processes. She was doubtful, however, that Biden and Xi would agree to AI-related nuclear arms control talks.
Breaking Defense reported that while the specifics of Biden and Xi’s AI-military talks were not publicly disclosed, with both sides releasing bare-bones press releases, there is a clear movement towards establishing norms on the military use of AI, particularly in autonomous weapons.
However, Breaking Defense said that neither side is likely to accept any ban limiting their freedom of action to deploy AI in their militaries.
Breaking Defense points out that US law prevents the President from making such a commitment without the approval of Congress and that China is seeking an expanded dialogue on AI rather than an agreement.
The same report says that while the US has been steadily building momentum to shape global AI norms, more than guidelines and political declarations are needed. Breaking Defense argues that the US should support clear and strong legal instruments that restrict the use of AI in autonomous weapons systems.
It also notes that China is thus far not following the US’s lead. Instead, Breaking Defense argues that China’s diplomatic strategy aims to counterbalance and rival the US in setting future AI standards, particularly in the military sphere.
While AI is a revolutionary technology, it also brings vulnerabilities. In a January 2022 Modern War Institute article, Nick Starck and other writers mention risks such as poisoning, evasion, reverse engineering and inference attacks.
Starck and other writers note that AI poisoning can occur due to adversaries altering training data, which can be countered by implementing rigorous data curation and governance policies.
They say that adversaries can use evasion techniques such as changing inputs to exploit the AI’s learned patterns, which can be prevented by evaluation tools and keeping humans in the loop.
They mention that adversaries can reverse-engineer AI by extracting its learned patterns, enabling them to predict or avoid responses, which can be stopped by strict accountability and access limitations.
In addition, Stark and others say that adversaries aim to deduce the data used for AI training in an inference attack, with that risk manageable by policy decisions regarding sensitive data.
Apart from those risks, the US may face a technological disadvantage with China as a near-peer competitor and a pacing challenge in AI technology.
The September 2022 Mid-Decade Challenges to National Competitiveness report by the Special Competitive Studies Project (SCSP) notes that the US has lost its edge in technology manufacturing, as much of high-end chip manufacturing is done in East Asia close to China. It says the US is now reliant on supply chains that are vulnerable to control by China.
The report also says that the US tech ecosystem overlooked geopolitical rivalries and the impact of tech advancements, leading to decreased US manufacturing capabilities. Consequently, the report says the US has strategic vulnerabilities due to a declining technology manufacturing base and a lack of proactive movements in technologies like AI.
In terms of policymaking, the report mentions that the US has been typically reactive instead of proactive in response to technological advancements, leading to strategic disadvantages. Those have been seen in America’s struggles to rebuild its microelectronics industry and prevent Chinese dominance in global 5G infrastructure.
It also points out that commercial priorities drive the US tech agenda due to the absence of a national effort equivalent to a modern “moonshot” and a decline in government-funded research and development (R&D) programs.
China takes a similar view regarding the military advantages and risks of AI. In an August 2020 Center for Security and Emerging Technology (CSET) report, Ryan Fedasiuk mentions that Chinese experts believe AI will enhance military capabilities through machine learning in tracking forces, lowering deployment costs and improving detection, targeting, and strikes against military targets.
At the same time, Fedasiuk claims that Chinese experts are worried that AI advancements in the military sphere could threaten strategic stability and reduce China’s defensive capabilities vis-a-vis the US.
He mentions Chinese experts’ tendency to overestimate the US military’s AI capabilities, leading to a push for increased investment in AI in China, akin to an arms race.
He mentions significant challenges to China’s AI development such as insufficient data collection, a lack of technical expertise and availability of advanced computing resources, and concerns about the leak and spread of its own AI advancements to the US.