Seven experts held a roundtable discussion on how to promote the "usability" and "trustworthiness" of large models.
Wu Hequan (Academician of Chinese Academy of Engineering)
Junlong Chen (Academician of the European Academy of Sciences, Deputy Director of Pazhou Laboratory, Dean of School of Computer Science and Engineering, South China University of Technology)
Wang Weiqiang (Chief Scientist of Ant Group Security Lab, Head of Trusted AI)
Chao Li (Chief Scientist of Tencent Cloud Tianyu)
In 2023, AIGC will sweep the world, and how to promote the safe development of artificial intelligence has attracted much attention.
On December 19th, the "Large Model Innovation Application and Security Governance Conference" with the theme of "Momentum AI Intelligence Enlightens the New Territories" was held in Guangzhou, which was guided by Nanfang Press Media Group and hosted by Southern Metropolis Daily and Nandu Big Data Research Institute in conjunction with Pazhou Laboratory and Guangming Laboratory. Relevant leaders of the Cyberspace Administration of the Guangdong Provincial Party Committee attended the conference.
At the meeting, Wu Hequan, academician of the Chinese Academy of Engineering, Chen Junlong, academician of the European Academy of Sciences and deputy director of Pazhou Laboratory, and other experts, as well as representatives from many universities, scientific research institutions, and industry enterprises gathered to share and discuss topics such as artificial intelligence large model technology, governance and application.
At the event, the Nandu Big Data Research Institute launched the "Leap Plan" and initiated the joint construction of the "AI Governance and Innovation Intelligent Alliance". Rong Mingchang, Secretary of the Party Committee and President of Southern Metropolis Daily, and representatives of relevant units and institutions witnessed the establishment of the "AI Governance and Innovation Intelligent Alliance".
Wu Hequan, Chen Junlong and other academicians talked about the big model
Discuss how to form safe, reliable, and controllable AI technology
In the theme sharing session, academicians, experts, and industry technical experts shared in-depth insights on topics such as "AI empowering the digital economy", "new data privacy", "large model security", and "large model construction, financial risk control".
Wu Hequan, academician of the Chinese Academy of Engineering, shared the theme of "Cloud Landing Large Model, New Chapter of Digital Economy Spectrum", he pointed out that the implementation of large models has given birth to a new round of technological innovation and industrial transformation, accelerated the empowerment of digital transformation, activated data elements to improve the quality of the digital economy, and enterprises that are good at applying data have become the main engine of economic growth. Moreover, with the embedding of inference capabilities with large models with tens of billions of parameters into mobile phones, intelligent assistant services such as simultaneous interpretation and intelligent map recognition are enriching our digital and intelligent life.
He emphasized that the implementation of large models on the cloud can accelerate the formation of new quality productivity and promote the quality and shift of the digital economy. Guangdong is a major manufacturing province, and it is hoped that Guangdong's basic model can be better integrated with the industry.
Chen Junlong, academician of the European Academy of Sciences, deputy director of Pazhou Laboratory, and dean of the School of Computer Science and Engineering of South China University of Technology, delivered a keynote speech on "Discussion on Data Intelligence and New Data Privacy". He emphasized that privacy and security must be considered in data sharing, and that in addition to the four privacy-preserving machine Xi methods, homomorphic encryption, differential privacy, federated Xi, and secure multi-party computing, the width Xi is also widely used at home and abroad.
With the topic of "The Urgency of Large Model Security and Practice Sharing", Wang Weiqiang, Chief Scientist and Head of Trusted AI at Ant Group's Security Lab, believes that safe, reliable, and controllable AI technology is urgently needed in the era Xi of large models, and the core is how to solve the safe, reliable, and controllable paradigm of generative learning in ultra-large-scale data. There are three key links to ensure the security of large models: one is to disinfect the training data source, the second is to strengthen the guardrail to prevent the uncontrollable problem of the black box in the inference process, and the third is to face externally induced adversarial attacks.
Li Chao, chief scientist of Tencent Cloud Tianyu, focused on the "lifeline" of the financial industry - risk control, and interpreted the application scenarios of large models. He pointed out that in the context of the refinement of customer operations and the increasing complexity of black market fraud, the response strategy of financial risk control has gradually upgraded from the traditional relatively static strategy confrontation to a more agile "model confrontation". In his view, the automated modeling technology based on the risk control model can help financial institutions to agilely iterate the risk control model that not only has rich risk control knowledge precipitation, but also adapts to specific business scenarios, so as to achieve dynamic risk management.
10 units and institutions jointly build an "AI Governance and Innovation Intelligent Alliance"
Nandu's large-scale research report on "Guangdong to Guangdong AI" will be launched simultaneously
In order to further help Guangdong accelerate the construction of a leading place for general artificial intelligence industry innovation, we will jointly explore the innovation of artificial intelligence regulatory models. At the conference, initiated by Southern Metropolis Daily and Nandu Big Data Research Institute, the first batch of 10 platforms and institutions jointly built the "AI Governance and Innovation Intelligent Alliance", government, industry, academia, research and media, jointly built an artificial intelligence development ecology, and explored good strategies for artificial intelligence governance.
The first batch of members of the "AI Governance and Innovation Intelligent Alliance" are Southern Metropolis Daily, Nandu Big Data Research Institute, Pazhou Laboratory, Guangming Laboratory, Guangzhou Artificial Intelligence Public Computing Center, School of Artificial Intelligence of Sun Yat-sen University, School of Artificial Intelligence Industry of Jinan University, School of Network Communication of Guangdong University of Finance and Economics, Institute of Innovation and Development of Law of Shenzhen University, Key Laboratory of Frontier Interdisciplinary Science of Artificial Intelligence of Hong Kong University of Science and Technology (Guangzhou), and Tahota Shenzhen Institute of Digital Economy Law. Representatives of 10 platforms and institutions and relevant leaders participated in and witnessed the co-construction ceremony.
It is reported that the large-scale research report of "Guangdong to Guangdong AI" will also be launched at the same time, and will carry out research, observation, evaluation and discussion from the perspectives of industrial foundation, computing power capacity, application scenarios, talent reserves, security governance, etc., present innovative achievements, share experience needs, build an exchange network, and fully and actively disseminate.
Nandu Big Data Research Institute launched the "Leap Plan"
Focus on four areas of pioneering and enterprising
It is reported that this conference is the first event of the 2023 Southern Metropolis Daily Think Tank Product Release Week. With the theme of "Deep Integration, New Connection, and High Intelligence", this year's Think Tank Week held 7 offline activities in Guangzhou, Shenzhen, Beijing, and Shanghai from December 19th to 23rd.
In the spring of 2018, Nandu put forward the goal and vision of "being a first-class think tank media in China", and the "Nandu Big Data Research Institute" came into being. It is also the main platform and important engine for Nandu to create new products, new services and new values of think tank media, and constantly collaborate and co-govern and empower innovation. At the conference on the innovative application and security governance of large models, Nandu Big Data Research Institute launched the "Leap Plan", Wang Weiguo, member of the Party Committee of Nandu Metropolis Daily, deputy editor-in-chief, and executive director of Nandu Big Data Research Institute, introduced that Nandu Big Data Research Institute will further focus on "digital economy and data governance", and focus on forging ahead in four major areas, namely: promoting "digital literacy and civilization improvement", giving full play to the guidance function of intelligent media; participating in the "circulation and transformation of data elements", giving full play to the function of intelligent media services, and maintaining" Network Content and Data Security", giving full play to the function of intelligent media governance, promoting "artificial intelligence trustworthiness and inclusiveness", and giving full play to the function of intelligent media integration.
For the transformation of Nandu intelligent media, academician Wu Hequan believes that Nandu focuses on the development of digital economy and data governance, highlights the characteristics of the times, and sends out a strong voice of the times, especially at the moment when artificial intelligence technology is booming, Nandu firmly seizes the opportunity of the times, and has made contributions to the promotion of artificial intelligence applications and encouraging AI technology innovation, and has made intelligent media responsibilities. Dean Chen Junlong commented, "Nandu is deeply integrated into the development of artificial intelligence through multiple intelligent media services, and I am very much looking forward to it." AI security governance requires the power of the media!"
Launched the AIGC Joint Initiative for the Development of Responsible Finance
It covers three values and eight basic principles
The emergence of generative AI has become a powerful "catalyst" for the digital transformation of finance. At the meeting, Nanfang Metropolis Daily and Nandu Big Data Research Institute, together with nearly 20 universities, scientific research institutions, and industry enterprises, launched the "Joint Initiative for the Development of Responsible Financial AIGC", which covers three values and eight basic principles, aiming to promote financial institutions, technology companies and other parties to apply generative AI more responsibly and ethically, and polish the background of finance for the people.
The event also released the "Cutting-edge Artificial Intelligence Application Case of the Year" and "Digital Intelligence Finance Pioneer Case of the Year", and the cases of seven companies, including Ant Group, Du Xiaoman, Qifu Technology, Immediate Consumption, Lexin, Sina Digital, and Rong 360, stood out.
Big coffee voices
Wu Hequan (Academician of the Chinese Academy of Engineering): The landing of large models on the terminal can drive the development of Internet platforms and related services, as well as digital content and media, which will enrich the new economic form to new quality productivity and promote the quality and shift of the digital economy.
Chen Junlong (Academician of the European Academy of Sciences, Deputy Director of Pazhou Laboratory, Dean of the School of Computer Science and Engineering of South China University of Technology): Guangdong proposes to accelerate the construction of a leading place for general artificial intelligence industry innovation, and we must harness the AI model to empower thousands of industries and help the high-quality development of the digital and intelligent economy.
Wang Weiqiang (Chief Scientist and Head of Trusted AI at Ant Group's Security Lab): In the new AI era, information authenticity and new risks need to be solved urgently, and new platform responsibilities have also been ushered in. In the era of large models, safe, reliable, and controllable AI technology is urgently needed, and the core is how to solve the safe, reliable, and controllable paradigm of generative Xi learning in ultra-large-scale data.
Li Chao (Chief Scientist of Tencent Cloud Tianyu): The actual effect of implementation is an important evaluation dimension of the value of large models, and the financial industry is a natural testing field for the effectiveness of large models, and data-driven decision-making and model application have penetrated into the front, middle and back offices. Nowadays, leading financial institutions have begun to use large model technology to automate and iterate risk control models in the core risk control process of financial retail credit business, so as to achieve more refined risk quantitative identification and decision-making.
Roundtable discussion
To meet the challenges of the development of AI large models, experts discussed and jointly built a new ecosystem of security governance
With ChatGPT as a starting point, a warm "100-model war" is underway. What are the current challenges faced by AI models, and how to develop responsible AI models? At the conference, academic experts and industry representatives gathered together to discuss "Exploring 'AI for Good': How to Promote 'Usable' and 'Credible' Large Models".
The development of AI large models faces three challenges
The development of AI models has triggered a new round of technological change, but it is also facing many challenges.
Wei Lijian, executive director of Sun Yat-sen University-Shansi Enlightenment AI Innovation and Empowerment Joint Laboratory, believes that the current development of AI large models in mainland China is facing challenges in three aspects: insufficient independent research and development capabilities, insufficient training data raw materials, and shortage of computing power. Fourth, strengthen the research and development of domestic general-purpose GPUs to meet the challenges of technical security and economic security at the national level.
Yi Huaijiong, the moderator of the roundtable and a senior partner of Tahota (Shenzhen) Law Firm, pointed out that in the process of iteration of AI large models, there are inherent risks in the collection, use, and algorithm training of data, such as the possibility of violating data security regulations. In addition, as a tool, the AI model may be maliciously used as a criminal tool, and there is a risk of purposeful behavior.
According to Ye Weiping, a professor at the Law School of Shenzhen University and dean of the Institute of Innovation and Development of the Rule of Law of Shenzhen University, the development of artificial intelligence technology itself and its commercial application are unpredictable, but in the process of technological development, we are constantly understanding, accepting and mastering it, and at the same time, the law is also following up on the security and privacy protection of artificial intelligence. In general, the security mechanism and the legal mechanism can interact and move forward in tandem, so the AI model is also expected to be safe and controllable.
Industry, academia and research need deeper integration and support
AI models have led the digital and intelligent innovation of thousands of industries. How to better innovate the AI large model research of universities, laboratories and other institutions, transform the application results in a timely manner, and empower the development of the industry?
Jiang Wenhao, a researcher at Guangming Lab and the leader of the generative large model team, believes that industry, academia and research need to be combined at a deeper level. "In many cases, researchers do not have a detailed understanding of the needs of enterprises, or even understand the new needs, so they need to have a deeper understanding and communication with each other in order to effectively promote the exchange of industry, university and research. "In addition, the research and business application of AI large models are relatively costly, and they need more support in terms of policies, taxes, and talents.
Liu Wei, associate professor and master's supervisor of the School of Artificial Intelligence at Sun Yat-sen University, pointed out that universities should establish closer cooperative relations with local enterprises and scientific research institutions, such as sharing scientific research results through regular seminars and workshops, sharing market demand, and accelerating the transformation of scientific research results. For example, for Guangdong's characteristic industries, customized scientific research projects can be jointly carried out in advanced manufacturing, financial industry, new energy and other fields.
Work together to solve the security of large model applications
Representatives from enterprises shared their difficulties and experiences in ensuring the security and trustworthiness of AI large models in their respective vertical fields from the perspective of industry practices.
Xiao Bing, algorithm director of the Consumer Artificial Intelligence Research Institute, said that compared with the general security requirements of large models, the financial industry has specific requirements and compliance scenarios, one is the general privacy requirements of large models, and the other is the compliance scenarios of the financial industry, such as the age limit of the customer group and the red line of loan interest rates. In addition, in enterprise online applications, the inference problem of the large model itself is also difficult to solve. In this regard, immediate consumption is mainly from the three levels of data, model, and solution to ensure the security and availability of large models. In addition, through cooperation with the government and academia, Immediate Consumption also detects various vulnerabilities and enhances the security of large models.
Liu Zhenquan, head of the Jinshan Office Security Center, said that WPS mainly encountered two major problems in the large model: data security, instability and unexplainability of the large model. For data security, WPS mainly takes measures in three aspects: organizational management, technology, and system to monitor and control the entire data life cycle. For the instability and unexplainability of the model, WPS has built a set of AI gateways, introduced a content moderation mechanism to manage the model, and created a prompt word platform subdivided into product domains, based on the characteristics of different models, flexibly debug prompt words and parameters, and ensure that the output results of the model meet user expectations, and comply with ethical standards and laws and regulations.
While ensuring the vigorous development of artificial intelligence, it has become a consensus to not lose the bottom line of ethics, not break legal boundaries, practice responsibility, and jointly build a new ecosystem of AI security governance. Yi Huaijiong believes that Nandu, together with laboratories, universities, research institutions, etc., initiated the establishment of the "AI Governance and Innovation Intelligent Alliance", and jointly issued the "Responsible Artificial Intelligence Large Model Financial Application Initiative" with industry enterprises, which are very good practices and fully reflect Nandu's role as an intelligent media. He concluded, "We need to change from passive to active, take the initiative to act, take the initiative to co-govern, and share development." ”
A08-09
Producer: Rong Mingchang, Liu Jiangtao
Planner: Wang Weiguo, Zou Ying
Co-ordinator: Ling Huishan, Kwan Jianming
Written by: Nandu reporter Zhang Yuting, Xiong Runmiao, Yuan Jiongxian, Xie Xiaoqing, Tang Jingyi, Luo Yun, Li Yujia
Design: Yin Jielin, Yuan Yi
Producer: Nandu Big Data Research Institute