Home / Security / Baidu denies any ties to reported Chinese military training on its GenAI chatbot

Baidu denies any ties to reported Chinese military training on its GenAI chatbot

tank-blue-gettyimages-1345226231

R_Type/Getty Images

Baidu has refuted suggestions it has ties to scientists who say they are training a Chinese military system based on the tech giant’s generative artificial intelligence (GenAI) chatbot, Ernie Bot. 

Media reports emerged on Friday that a university linked to the People’s Liberation Army (PLA) of China was training the military’s AI system with Baidu’s large language model (LLM). Researchers from the Information Engineering University fed the LLM with data on weapons and forces deployed in past conflicts to predict combat strategy, the South China Morning Post first reported.

Also: China and US part of multilateral pact to collaborate on AI risks

Citing an academic paper published last month in the Chinese journal Command Control & Simulation, the SCMP report said other LLMs also were used, including GPT3.5 and GPT4. The LLM then was prompted to generate combat strategies and outcomes based on the training data. 

In the paper, the researchers referenced the US military invasion of Libya in 2011 and said they used combat tactics employed by both sides of the conflict, and input the desired targets for Ernie Bot to predict the US army’s strategy.

The research team said the simulated results could aid decision-making and be used to further train and enhance the LLM’s knowledge and cognitive understanding. The goal was to improve AI-powered military systems and enable them to better understand human intentions, which was critical as human adversaries can be unpredictable.

The researchers added that their work was in its early stage and carried out only for research. 

Baidu’s shares took a dive after reports on the research surfaced. 

The Chinese tech vendor since has refuted having any links to the authors of the paper or the university. In a statement referring to the academic paper, Baidu said Ernie Bot is available to and can be used by the general public. 

Also: The best AI chatbots: ChatGPT and other noteworthy alternatives

Baidu added that it had “not engaged in any business collaboration or provided any tailored service” to the team that penned the academic report or any institutions with which the researchers are affiliated. 

It noted that the original SCMP article has been amended to reflect its statement. 

Baidu said: “[We are] committed to operating its AI-related products and businesses in compliance with applicable laws and regulations and best corporate practices.”

The Chinese search engine and autonomous vehicle maker last August released Ernie Bot to the general public in hopes of gathering user feedback to improve the platform. It also plans to release AI-native apps that it said would deliver four key GenAI capabilities, encompassing reasoning, memory, generation, and understanding. 

Also: Most people support the need for trustworthy and regulated AI

Ernie Bot, among other touted capabilities, can generate text, images, audio, and video given a text prompt, and can deliver voice in several local Chinese dialects, including Sichuan. Since its introduction with limited access in March, the AI model has gained proficiency in more than 200 genres of writing, clocking a 1.6-fold increase in content quality, according to Baidu. Five plugins were also added to Ernie Bot this month, including Baidu Search, data analytics and visualization, and text-to-video.  

In November last year, China’s Defense Ministry urged nations to be “prudent and responsible” in their use of AI in military, according to a report by state-owned media China Daily. It noted that the Chinese government paid close attention to security risks posed by AI in military applications. “China is opposed to using AI advantages to undermine the sovereignty of other countries,” said the Defense Ministry’s spokesperson Wu Qian.

In other news this week, China’s third-generation quantum computer Origin Wukong has processed 33,871 tasks for users worldwide since it became operational on January 6, reported state-run media agency Xinhua

Users from more than 60 countries remotely accessed the quantum computer at least 350,000 times, with those from the US leading the pack in terms of frequency, according to the Anhui Quantum Computing Engineering Research Center. The quantum computer runs on a 72-qubit local-made chip. 


Source link

About admin

Check Also

Gen AI training costs soar yet risks are poorly measured, says Stanford AI report

The number of significant new AI models coming out of industry has surged in recent ...

Leave a Reply

Your email address will not be published. Required fields are marked *