How to build a "safety embankment"?, Faced with the "wave" of new technologies

Since this year, important events such as the Digital China Construction Summit, the China Cyber Civilization Conference, and the 2023 National Cyber Security Publicity Week have been held in Fujian. These three events are all related to the Internet, but they focus on two different keywords.

The Internet is a "double-edged sword", and "development" and "security" are both sides of it. With the deepening development of a new round of technological revolution and industrial transformation, the risks and challenges facing the field of network security are becoming increasingly prominent.

For example, in recent years, the "AI face swapping" technique, which has received increasing public attention, also received attention for its "attack" and "defense" at this year's cybersecurity expo.

With just a photo of the target person and artificial intelligence synthesis, standing in front of the camera, you can "turn" into him on the screen. You can also pass facial authentication on the online lending platform and transfer money with just a blink of an eye and a shake of your head. At the Meiya Baike exhibition area, staff demonstrated this "scam".

Regarding this, on-site staff stated that currently, Meiya Baike has independently developed a deep forgery detection and identification platform, which can intelligently detect and identify forged videos, helping regulatory authorities improve their content security governance capabilities.

When discussing "development" at the Digital China Construction Summit held in Fuzhou in April this year, AI was arguably the hottest term; When discussing "security" during the current Cybersecurity Awareness Week, AI remains the biggest focus.

"Data can be said to be the most important element of AI big models, and everyone has different understandings, but overall, almost everyone believes that the quality of data has the highest impact on the intelligence level of the model, about 60%." In the view of Huang Tiejun, Dean of Beijing Zhiyuan Artificial Intelligence Research Institute and Professor of Beijing University School of Computer Science, big models are based on the data behind them, and all functions come from data, which also means that the risk is increasing. "For AI technology, if you give it 'wrong' data, it will always follow this wrong approach, sometimes contradictory to each other."

Shen Xiangyang, the founding director of the Guangdong Hong Kong Macao Greater Bay Area Digital Economy Research Institute, used three words when talking about the impact of AI on society: imperfect data sources, serious "nonsense", and even harder to detect AI rumors.

Both search engines and AI models have the function of collecting information and answering questions. Shen Xiangyang vividly describes search engines as "limited liability companies" and AI models as "unlimited liability companies". The difference in this statement lies in deep learning.

"When search engines find something bad, you know where it comes from, so its responsibility is limited. However, the AI big model generates content through deep artificial intelligence, and I call it the 'Responsibility Unlimited Company'," said Shen Xiangyang.

In the view of some experts, generative artificial intelligence and new security risks are emerging, such as negative public opinion, algorithmic bias, cognitive errors, psychological misleading, data leakage, etc.

In July of this year, China introduced the world's first specialized legislation for generative artificial intelligence - the "Interim Measures for the Management of Generative Artificial Intelligence Services", which put forward clear requirements for data compliance.

Zhuang Rongwen, Deputy Minister of the Central Propaganda Department, Director of the Central Cyberspace Administration, and Director of the National Cyberspace Administration, stated at the meeting that we need to innovate data governance, vigorously improve data security and personal information protection levels, increase the crackdown on illegal and irregular activities in the data field, and ensure personal information security. We need to strengthen regulatory guidance, conduct forward-looking research on new technologies and applications, draw a clear safety red line, and effectively regulate and guide the healthy, orderly, and safe development of new technologies and applications.