Chinese tech firm Baidu is likely to face sanctions from the US as a result of reports linking its artificial intelligence (AI)-driven ‘Ernie’ chatbot to China’s People’s Liberation Army (PLA), analysts have suggested.
On Monday (15 January), Hong Kong newspaper the South China Morning Post cited an academic paper claiming the PLA had tested its large language model (LLM) on Baidu’s ‘Ernie’ bot and on AI firm iFlytek’s Spark.
The paper came from a Chinese university affiliated with the PLA Strategic Support Force, described as a “new branch of China’s military” by Wilson Jones, a defence analyst at GlobalData.
“The Strategic Support Force focuses on electronic, cyber, and space warfare, but military AI development falls firmly under its mission”, Jones told Verdict.
Following the report from the South China Morning Post, Baidu’s shares slumped by 14.7% to 96.8 HKD ($12.4) today (17 January), its lowest point in more than a year.
Internet users in Hong Kong and Taiwan reached a ‘404’ error page when trying to access an online link to the institute’s research paper on the Baidu-PLA affiliation.
The paper was still accessible from mainland China, Bloomberg reported.
Beijing hedges military bets on AI
Baidu, which said it had “no knowledge of the research project, and if our large language model was used, it would have been the version publicly available online”, has frequently been touted as China’s answer to OpenAI.
After making the Ernie bot publicly available in August 2023, Baidu said in December that it had garnered more than 100 million users.
The tech company has become China’s leading AI developer in recent months, remaining ahead of larger rivals such as Tencent.
Baidu’s unprecedented growth has come as the Chinese Government invests vast amounts in military tech.
In 2023, Beijing boasted the second-largest defence budget in the world at $230.3bn, according to a GlobalData report, while the PLA regularly funds “university scholarships for military technology students, including for AI tech”, Jones said.
China is far from the only nation with considerable entanglement between private tech firms and military institutions, seen by the growing dominance of SpaceX’s Starlink satellites within the US’ space missions.
Regardless, the US response “will probably be sanctions and a ban on Baidu products for elected politicians and military service members”, Jones added.
A similar approach was taken in 2022 when Washington banned telecoms products from Huawei due to concerns over data collection and mass surveillance.
AI or human soldiers?
LLM testing aside, AI military adoption is becoming increasingly prevalent on the battlefield.
Combat scenarios from Ukraine to the Middle East have been flooded with loitering munitions, which hover above a designated warzone and identify enemy targets before striking.
The Shahed-131 and Shahed-136 are known as the ‘suicide drones’ because they fly into targets and explode on impact.
Ukraine and the US have previously accused Tehran of supplying Moscow with hundreds of Shahed UAVs through production sites in Belarus and Russia.
The morality of automatising the decision to strike is highly controversial.
In most war zones, personnel are taught the ‘OODA Loop’ (Observe, Orient, Decide, Act) when processing surroundings and responding to enemy action.
“AI computers can calculate these processes much faster than a human brain ever could," Jones said. “For example, an AI air defence system could identify a flying target, calculate its speed and trajectory, determine if it is friend or foe, and decide how to intercept in a matter of minutes.”
