[Quick Review of the Two Sessions]

Zhang Mengran

At this year’s National Two Sessions, Zhang Tianren, deputy to the National People’s Congress and chairman of Tianneng Holding Group, suggested that an Artificial Intelligence Management Law should be formulated to address a series of socioeconomic, moral and ethical issues that artificial intelligence (AI) may cause. The first priority is to establish a complete ethical system for artificial intelligence technology. Chen Xiaohong, a member of the National Committee of the Chinese People's Political Consultative Conference and an academician of the Chinese Academy of Engineering, also said that when actively building an open source innovative AI ecosystem, it is necessary to promote the sharing of computing resources and model iteration collaboration, and to improve the AI ​​ethical value system.

Today, AI technology is sweeping in like a raging wave, and its powerful functions have brought unprecedented development opportunities to human society. However, behind the booming development, a series of ethical governance problems lie like hidden reefs. Among them, AI authenticity and AI values ​​​​are two aspects that cannot be ignored.

Among the authenticity challenges accompanying the development of AI, the issue that is closely related to people is the issue of false information. For example, deep forgery technology based on generative adversarial networks can generate highly realistic false videos, which may harm personal interests at best, or affect social order and public safety at worst.

To deal with this challenge, on the one hand, developers should incorporate transparency and verifiability mechanisms into every aspect of model training and algorithm deployment to prevent the generation and spread of false information from the source; on the other hand, the introduction of relevant laws and regulations should be accelerated to punish the use of AI to create and spread false information.

In addition to the authenticity problem, AI values ​​​​are also a top priority in ethical governance. We know that the learning of AI systems is based on massive data sets, which often contain various biases of human society. If biases in real society are not effectively corrected, they are likely to be further amplified after being learned by the AI ​​system.

A typical example is that some Western countries’ AI facial recognition systems have abnormally high misjudgment rates when used for ethnic minority groups. Data shows that when identifying people of color, several systems based on deep learning technology have error rates as high as 20% to 30%, while the error rates when identifying white people are much lower. The formation of this "AI bias" is, at the technical level, due to the serious racial inequality in the original database samples used by the developers. This might even lead algorithms to associate dark skin features with "higher risk." Once such facial recognition systems are used in the financial, security and law enforcement fields, it may lead to unfair treatment.

Therefore, from the perspective of ethical governance, developers should introduce diversity and diverse perspectives during the data collection, model training and algorithm design stages to avoid monoculture or group bias dominating AI. In addition, governments and industry organizations should use review mechanisms to ensure that AI maximizes compliance with basic social ethics, or encourage interdisciplinary cooperation to allow legal experts, ethicists, and social scientists to participate in the ethical assessment of AI.

Technological innovation is undoubtedly a powerful driving force for the development of AI, but at the same time, it is also necessary to equip the technological giant AI with reliable navigation and escort mechanisms. In this process, every participant, whether they are technology developers, enterprises or policy makers, shoulders a common responsibility.