bodyType

customContents

entityId

entityUuid

articlePageQuery

filter

article

articleBaseAdvertisingInfoArticle

articleListArticle

articleSeoArticle

helpersCheckIsPostiesArticle

hooksArticleAcknowledgementGateOverlayArticle

hooksTrackPageViewArticle

hooksContentInterestArticle

articleLinkingDataContent

articleSchemaContent

liveArticle

advertZone

sentiment

contentLock

topics

types

sections

moreOnThisArticles

urlAlias

commentCount

sponsorType

authorLocations

authors

paywallTypes

corrections

displaySlideShow

multimediaEmbed

printHeadline

seriesContainer

socialHeadline

headline

flag

firstTopic

glossary

flattenTypeIds

publishedDate

additionalInfoBoxes

images

image

relatedLinks

relatedNewsletters

__typename

createdDate

sponsor

factSheets

series

knowledgeQuestion

disableOffPlatformContent

published

updatedDate

summary

subHeadline

body

keywords

copyrighted

customTimezone

liveContents

Stephen Chen

A generative AI that can boost the performance of electronic warfare drones has been developed by scientists in China’s defence industry.
This large language model (LLM), similar to ChatGPT, can command a drone equipped with electronic warfare weapons to attack enemy aircraft radar or communication systems.
Its decision-making performance in air combat not only surpasses traditional artificial intelligence (AI) techniques such as reinforcement learning, but is also better than experienced human experts, test results suggest.
This is the first publicly disclosed research that directly applies large language models to weapons.
Previously, such AI technology was mainly limited to war rooms, offering intelligence analysis or decision-making help to human commanders.
The research project was jointly launched by the Chengdu Aircraft Design Institute under the Aviation Industry Corporation of China and Northwestern Polytechnical University in Xian, Shaanxi province.
The institute is the designer of China’s heavy stealth fighter J-20, while the university has been in the crosshairs of cyberattacks launched by the US government, owing to its pioneering research in cutting-edge technology.
According to a paper published by the project team on October 24 in the peer-reviewed Journal of Detection & Control, the work is still in the experimental stage.
Among existing artificial intelligence technologies, LLMs are the most adept at understanding human language.
The project team fed the LLM with a wealth of sources, the team said, including “book series on radar, electronic warfare and related literature collections”.
More sensitive materials, including air combat records, weapons inventory set-up records and electronic warfare operation manuals, were also put to the model.
Most of the training materials were in Chinese, according to the researchers.
Electronic warfare mirrors a cat-and-mouse game, where the attacker unleashes specific electromagnetic waves to quell the radar signals emitted by the target.
The defender, in turn, tries to evade these attacks by continuously altering signals, forcing the opponent to adapt its strategy in real time based on surveillance data.
Previously, it was believed that LLMs were ill-suited for such tasks due to their inability to interpret sensor-collected data.

Generative AI also often demands longer contemplation times, falling short of the millisecond-level response speed essential in electronic warfare.
To circumvent these challenges, scientists delegated the raw data processing to a less complex reinforcement learning model. This traditional AI algorithm excels in comprehending and analysing vast amounts of numerical data.
The “observation value vector parameters” extracted from this preliminary process are then transformed into human language via a machine translator. The large language model then takes the reins, processing and analysing this information.
The translator converts the large model’s responses into output instructions, which ultimately control the electronic warfare jammer.
Test results have confirmed the feasibility of this technology, according to the researchers. With the aid of reinforcement learning algorithms, generative AI can swiftly adjust attack strategies up to 10 times per second.
When pitted against traditional AI and human expertise, LLMs prove superior in creating a multitude of false targets on enemy radar screens. This disruptive strategy is deemed more valuable in the realm of electronic warfare than mere suppression with noise or deflecting radar waves away from genuine targets.


“They still have many practical issues to address, including chips, the size of model and security risks,” said a Beijing-based AI scientist who requested anonymity due to the sensitivity of the issue.
“But there’s little doubt that ‘words can kill’ is evolving from a philosophical concept to reality,” he said.


J-20 designer arms electronic warfare drones with generative AI

China voices support for Nasa over botched Boeing Starliner space mission

Why the year AD1 was also a new era for ancient China

China floats wave, wind and solar power plan for South China Sea outposts

Science

Chinese scientists have developed generative AI that can command drones in electronic warfare, according to a recent paper. Photo: Shutterstock

The designer of the Chinese J-20 stealth fighter is part of the research team involved in the AI project. Photo: Weibo

China’s Zhipu AI says its app can operate your smartphone for you

People look at their smartphones outside a restaurant in Shanghai. Photo: AFP

Zhipu AI is one of China’s top start-ups working on generative AI. Photo: Handout

Zhipu AI’s AutoGLM can complete certain tasks through a select group of apps, such as WeChat and Meituan. Photo: Screenshot 

OpenAI, Broadcom working to develop AI inference chip

A smartphone with a displayed Broadcom logo is placed on a computer motherboard in this illustration taken March 6, 2023. Photo: Reuters

OpenAI CEO Sam Altman speaks during the Microsoft Build conference in Seattle on May 21, 2024. Photo: AFP/Getty Images/TNS

From physics to mind: where AI really came from

Frank Wilczek. Photo: Niklas Björling at Stockholm University

The 2024 Nobel Prize for Physics was awarded to AI pioneers Professor John Hopfield (left) of Princeton University, and Professor Geoffrey Hinton, of the University of Toronto. Photo: Princeton University via AP

OpenAI chair’s AI start-up Sierra gets US$4.5 billion valuation in latest funding round

Artificial intelligence letters are placed on a computer motherboard in this illustration taken, June 23, 2023. Photo: Reuters

The OpenAI and ChatGPT logos are seen in this illustration taken February 3, 2023. Photo: Reuters

US to ban outbound investments in China’s chips, AI over national security

US President Joe Biden exits Marine One as he arrives in Washington on Monday. The American leader’s term in office ends in January. Photo: AP

A view of a facility belonging to Semiconductor Manufacturing International Corporation, also known as SMIC, in Pudong district in Shanghai. SMIC is mainland China’s largest semiconductor foundry company. Photo: AFP

News

China

Chinese scientists have developed a large language model that can command military drones to attack enemy radar systems


Chinese scientists have developed a large language model that can command military drones to attack enemy radar systems.


Artificial intelligence

Drones

China’s military

China technology