Have you ever experienced such a "weird" thing? Just browsed a few pages of a qualification exam website, the training institution called; last week just hung the house you wanted to sell on an intermediary website, the next week never contacted several intermediary companies have come to the door; the photo I accidentally uploaded to the social networking site, not only was it stolen inexplicably, but even used to illegally benefit ... While we are immersed in the convenient life brought by AI technology, problems such as uncontrollable technology, data security, and privacy leakage caused by the "black box mode" occur from time to time, which makes people worry.
On December 14, the first phase of the first quarter of the "Talking about Cloud Says AI" Technology Industry Salon was launched, and in response to the topic of trusted AI, which is widely discussed at the moment, He Fengxiang, algorithm scientist of Jingdong Exploration Research Institute, Chen Junyan, director of the Artificial Intelligence and Big Data Division of the East China Branch of China Academy of Information and Communications Technology, and Wang Yunhe, head of standards and strategic director of Huakong Qingjiao Information Technology (Beijing) Co., Ltd., gathered together to discuss in depth "how to balance the credibility and availability of AI technology". Advice for trusted AI on the spot.
The frequent crisis of trust AI technology has led to more thinking
At the beginning of this year, the pony who lives in Shanghai encountered such a troublesome thing: just downloaded a game and performed face recognition according to the requirements of the game, how could it be that a month later, the pony received his credit card swipe record in Hainan, which was simply shocking...
The occurrence of the above incidents is no accident, accompanied by the emergence of AI credibility problems represented by personal privacy leakage and data abuse, which has indeed brought many adverse effects to social stability and personal development. "Nowadays, there are many sources of information and it is impossible to trace the source in a short period of time, and if there is a regulatory loophole in data use, it will directly lead to personal privacy leakage and even direct financial losses." Chen Junyan further analyzed the current trust crisis in AI technology.
Chen Junyan, Director of the Artificial Intelligence and Big Data Division of the East China Branch of the China Academy of Information and Communications Technology
The AI trust crisis may also trigger more serious personal safety issues. "Because AI technology is based on a black box model, in some key areas, such as medical diagnosis, autonomous driving, and other life-threatening industries, if there is a crisis of trust, it will have very terrible consequences." He Fengxiang analyzed that these crises will have a huge impact on the development of the industry and the landing of applications.
What is more serious is that if a crisis of trust occurs, it is difficult to define and pursue the rights and responsibilities, and even cause social chaos. Relevant information shows that since the beginning of this year, a total of 8 various licensed enterprises have accumulated 89 autonomous driving traffic accidents. "Today's autonomous driving accidents are the responsibilities of algorithms, human responsibilities, or the inaccuracies of identification or the uncertainty of the road environment, it is difficult to define." Wang Yunhe analyzed, "For example, in AI applications, some people provide data modeling, some people provide models, if the model is trained once there is an error, because of multi-party participation, it is difficult to define where the problem is." ”
Multi-party efforts to explore and work together to build a trusted AI ecosystem
In the face of the above-mentioned crisis of trust, all walks of life have been looking for solutions. Chen Junyan said that in view of the implementation of trusted AI, the international community mainly regulates the credibility of artificial intelligence in the form of guidelines and bills, with ethics and people-oriented as the core of trusted AI; there are also cases that are constantly explored in the formulation of standards, and through the construction of artificial intelligence management framework, trusted AI is supervised to enhance trust. "But there are some imperfections in these areas, such as bills or guidelines that will make trusted AI have clear boundaries, but it is very easy to have a 'one-size-fits-all' situation." In contrast, unlike foreign countries, China continues to improve the policy level, from the policy, ecology, law, standard formulation and other aspects of comprehensive efforts, such as the "National New Generation of Artificial Intelligence Standard System Construction Guide", "Promote the Development of Trusted Artificial Intelligence Initiative" and other documents.
In this regard, the JD Exploration Institute is increasing its research on trusted AI from four major aspects, namely stability, interpretability, privacy protection and fairness. He Fengxiang said: "We start from the basic theory to study the most cutting-edge problems, start from these theoretical problems and theoretical results, design new and trustworthy AI algorithms, and then try to apply these AI algorithms to products to complete the final landing of technology." At the exchange site, he also shared the exploration of Jingdong Exploration Research Institute in trusted AI, and said that he very much hopes to participate in the formulation of standards to further ensure the credibility of AI in the future.
He Fengxiang, algorithm scientist at Jingdong Exploration Research Institute
In addition, Wang Yunhe also proposed: "In the process of promoting the construction of trusted AI, the AI industry needs the joint efforts of all parties. From the perspective of data circulation and data privacy protection, cryptography technology represented by multi-party secure computing can be added to consolidate the underlying support capabilities of data circulation infrastructure and accelerate the landing of standards and industries. ”
Wang Yunhe, head of standards and director of strategy of Huakong Qingjiao Information Technology (Beijing) Co., Ltd
For how to solve the problem of AI trust and develop a virtuous circle in the future of AI, the three industry leaders have invariably given the ultimate answer: the development of AI trust ecology. "Trusted AI involves a wide range of practical tasks, and it is difficult for an enterprise or institution to bring about qualitative creation and leaps by its own efforts, so the whole chain of industry-university-research collaboration is the key; in the future, the industry needs to build a mature and open ecosystem, and the strength of the crowd can bring more breakthroughs in technology, standards and practices." On behalf of the JD Exploration Research Institute, He Fengxiang expressed his imagination for the future of trusted AI.
"From the industrial level, first of all, we must build a legal system and improve industry standards, the two complement each other and are indispensable; secondly, we must unite all parties to practice trusted artificial intelligence." In this process, enterprises are actually the backbone of the practice of trusted AI, and can solve problems such as privacy leakage, algorithm bias, and content review in data screening, algorithm optimization, and model design to promote AI technology to be more credible. Chen Junyan focused on it.
As the salon drew to a close, the three experts further elaborated on other key issues of trusted AI in response to questions from the audience. It is worth affirming that this salon provides many useful references for the future development of AI technology through the discussion of the availability and credibility of AI, and brings impetus to the in-depth exploration of the landing of trusted AI and the construction of a trusted ecology.
—END—