Revise Chinese military university’s AI course to ‘educate for war’: researchers

South China Morning Post | Meredith Chen

Researchers at China’s Central Military Commission-overseen National University of Defence Technology have proposed redesigning their graduate AI curriculum to embed real-world combat scenarios and train officers directly for intelligent warfare. Outlined in the July issue of Defence Industry Conversion in China, the proposed course restructures technical instruction around uncrewed warfare operational stages, including situation assessment, automated targeting, and strike evaluation.

The initiative reflects Beijing's effort to integrate emerging technologies into military command decisions. Practical modules incorporate war gaming for reinforcement learning and fault diagnosis in uncrewed systems during simulated border intrusions. Human oversight remains mandatory. Students will use a custom online simulation platform to conduct multi-agent drone swarming exercises, competing in mock combat to develop tactical AI algorithms. Introduced in autumn 2022 for non-AI engineering graduates, the updated framework explicitly maintains that AI acts strictly as decision support, leaving command authority entirely with human officers.

Comment

Embedding AI decision-support platforms within Central Military Commission-mandated command structures creates an operational tension between algorithmic speed and manual authorisation. By requiring graduate researchers at the National University of Defence Technology to enforce strict human control over drone swarms, Beijing codifies a conservative command-and-control framework. This doctrine ensures that human commanders retain ultimate weapon-release authority during sensor-to-shooter cycles. However, processing real-time sensor streams from autonomous systems threatens to overwhelm staff capacity in saturated combat environments.

Consequently, this human-in-the-loop requirement introduces a severe latency bottleneck into NUDT-modelled autonomous swarm engagements. Future PLA Theatre Command operational doctrine will likely resolve this bottleneck by establishing pre-approved execution thresholds for algorithmic strike vectors. That adjustment shifts human oversight from real-time tactical control to post-strike assessment inside PLA Joint Operations Command Centres.

Strategic Question for Discussion

If PLA Joint Operations Command Centres enforce strict human authorisation during multi-agent drone swarm operations, does the resulting decision latency negate the speed advantages of algorithmic warfare, or can pre-approved strike parameters preserve tactical momentum?

My assessment is that rigid human-in-the-loop requirements will inevitably create critical bottlenecks during high-density engagements, forcing commanders to delegate pre-authorised execution boundaries to algorithmic agents. The available evidence suggests that PLA operational doctrine will adapt by reserving human intervention for strategic target selection while allowing autonomous platforms to execute tactical swarm manoeuvres independently.

Share your assessment in the comments below.

To read in details click here