On September 27th, the artificial intelligence sub-forum of the 2021 World Internet Congress was held in Wuzhen. Mei Tao, vice president of JD.com Group and vice president of JD.com Institute of artificial Intelligence, was invited to attend the peak dialogue of "Building a credible New Future of AI". Mei Tao said that the current widespread landing of artificial intelligence is the best time to discuss trusted AI. JD.com is committed to creating reliable AI products and services with excellent technology, coordinated development and customer trust based on the standards of stability, interpretability, privacy protection and fairness, so as to promote the high-quality development of the digital economy and speed up the construction of digital China.
Mei Tao, vice president of JD.com Group and vice president of JD.com artificial Intelligence Research Institute.
At present, with the deepening of the integration of technology and real economy, new models and new business type emerge in endlessly, and the long-term sustainable and healthy development of technology requires the construction of a good technological concept. In the field of artificial intelligence, "trusted AI" has become the necessity of its development and the focus of the industry and society.
Mei Tao said that after a lot of exploration and practice, JD.com proposed a set of AI credible evaluation system, which includes stability, interpretability, privacy protection, fairness and other standards. Specifically, stability means that the model and algorithm of artificial intelligence can resist external malicious attacks and ensure the stability and security of the model; interpretability means that the decisions made by artificial intelligence should be explained and understood by human beings; privacy protection means that artificial intelligence can protect the privacy of information; fairness means that artificial intelligence treats all users fairly, so that everyone can enjoy the technology dividend without difference. In Dr. Mei Tao's view, if artificial intelligence reaches a high level in the above four metrics, it will be able to be clear, transparent and credible.
JD.com also implements these four standards all the time in the process of product research and development and landing. According to Dr. Mei Tao, JD.com has carried out a series of industrial practices in the intelligent manufacturing scene, polishing the intelligent manufacturing industry quality inspection mechanism based on zero sample or small sample learning, and its model has better generalization. In the deployment of the model, through its own privacy training platform to ensure information security in the deployment process. In the intelligent circulation link, in order to ensure the versatility of the model, a 199-layer three-dimensional neural network is designed, which can identify whether the operator's operation behavior in the process of logistics is standardized or not. in this way, the generalization and versatility of the large-scale pre-training model can be improved.
In the field of intelligent customer service, JD.com uses a multi-modal way to understand users' emotions and behavior in the working process of intelligent customer service, and creates a lot of cross-modal artificial intelligence models. As the first batch of enterprises to pass the certification of "Digital trusted Service Evaluation" of China Information and Communication Institute, its products such as "full scene service" and "secure and trusted" are highly recognized by the experts of the jury. It is reflected in stability, interpretability, privacy protection and fairness. For example, for fairness, JD.com intelligent customer service has specially developed accent adaptation models and algorithms to enable it to recognize each other's dialects (such as Sichuan and Chongqing, Shaanxi, Shanxi and other dialects). So that more people can enjoy the convenience of JD.com 's intelligent customer service.
In fact, as early as the Internet Conference in Wuzhen in 2019, JD.com took the lead in systematically expounding the concept and far-reaching significance of reliable AI in China. This year, JD.com Exploration Research Institute and China Information and Communication Institute officially released China's first trusted artificial Intelligence White Paper at the World artificial Intelligence Conference, which systematically put forward a panoramic framework of trusted artificial intelligence. this paper comprehensively expounds the characteristic elements of trusted artificial intelligence, and makes a comprehensive analysis of the relationship between trusted artificial intelligence and artificial intelligence science and technology ethics and governance, in order to promote the industry to better understand trusted artificial intelligence.
JD.com has devoted himself to the standard exploration and practice of trusted AI for many years, and has continuously promoted the development of trusted AI in practice. Mei Tao believes that in the future, the credibility of AI will be a key difficulty for the landing of artificial intelligence, and only products that follow trusted AI specifications and ensure stability and security can win the long-term trust of customers.