Double Sided AI: Risk And Credible Game, How Can Science And Technology Be Good?

From autonomous driving, AI healthcare to industrial production, as well as all areas of our lives, AI is showing the potential for tremendous change empowerment. However, with the development of AI application, many problems, such as algorithm security, data discrimination, data abuse and so on, also show some risks and hidden dangers. As a result, people think about the trust of artificial intelligence.

On July 9, the initiative to promote the development of trusted artificial intelligence (hereinafter referred to as the "initiative") was officially released at the 2021 World Conference on artificial intelligence's trusted AI forum. The initiative was initiated by the trusted AI forum and jointly released by the Chinese Academy of communications, the China UK artificial intelligence ethics and governance research center of the Institute of automation, Chinese Academy of Sciences, Jingdong Exploration Institute, ant group, etc.

The above initiatives aim to promote the formation of international consensus on collaborative governance, promote the understanding and attention of the community on trusted AI, and fully tap the industrial and social value of AI. Wu Jincheng, director of the Shanghai Economic and Information Commission, told reporters of the 21st century economic report that developing credible AI, promoting the transparency, visibility, safety and controllability of AI, and diversity and inclusiveness have become the global consensus and the only way for the healthy development of AI in the future.

He believes that there are three ways to promote the application of encryption technology, secure multi-party computing, federated learning and other security technologies. Improve the security features and data protection capabilities of AI technology. The second is to establish a test and evaluation system. Accelerate the construction and improvement of trusted AI related standards, and develop innovative testing and evaluation technology. The third is to speed up the improvement of system norms. Under the framework of relevant national laws and local regulations of Shanghai, laws and regulations in the fields of bio information security, data security and personal privacy are studied and formulated to promote the development of trusted AI with institutional guarantee.

From autonomous driving, AI healthcare to industrial production, as well as all areas of our lives, AI is showing the potential for tremendous change empowerment. Gan Jun system

Risk control is no longer a human struggle

In fact, when we give decisions to AI, AI must first prove that it is trustworthy. Therefore, the study of trustworthiness based on this has formed a global consensus.

Xu Peng, director of the high technology division of the science and Technology Department of the Ministry of industry and information technology, said that risk prevention and challenges in the security field, especially those closely related to human production and life, such as ethics, algorithms, abuse of personal information and disclosure of privacy, as well as corresponding protection issues, are now one of the hot topics of concern. Therefore, it is very urgent to construct the basic security system of AI and promote the healthy development of AI.

"Risk control in the future will no longer be a struggle between people, but a game between AI and AI." Dr. Qi Yuan, vice president of ant group and chief artificial intelligence scientist, told reporters of 21st century economic report that on Alipay, behind hundreds of millions of transactions every day, it is not retail investors who launch attacks, but professional gangs of illegal production.

In order to deal with the complex and changeable black industry attack, how to continuously strengthen the anti attack ability of the risk control model, so as to achieve "the devil is one foot high, the road is ten feet high", has become a difficult problem for all AI practitioners.

It is understood that since 2018, Qiyuan and a group of AI technical backbones have integrated the economic "game theory" into the artificial intelligence model algorithm, and developed a brand-new industry technology - "intelligent confrontation", that is, by combining AI and game confrontation, the machine can automatically, real-time, and dynamically attack its own system, so as to improve the model ability, It makes the model processing result more accurate.

This technique is similar to "fighting with hands" in Jin Yong's novels. Because of its ability to form perception, analysis and prediction of risk knowledge, it will become the core security technology in the field of trusted AI in the future.

According to Zeng Yi, director of the Sino British artificial intelligence ethics and governance research center of the Institute of automation, Chinese Academy of Sciences, a credible artificial intelligence system must have online continuous learning and correction ability at any time“ From a technical point of view, the latest models of artificial intelligence technology progress are not credible now. Now the technology of artificial intelligence is not enough to talk about real credibility, so it is a starting point. In addition, trusted artificial intelligence is definitely not only a technical problem, but also a social perspective. How to use technology to empower cross-cultural mutual trust, in my opinion, is what we have to do to truly realize credible artificial intelligence, and even the key to the final success. "

Urgent need to establish industry standards

As the key to "resist risks" in the digital economy, trusted AI technology has been implemented in many risk control scenarios of ant group, including anti fraud, anti money laundering, anti gambling, joint risk control of enterprises and data privacy protection.

In addition to financial risk control scenarios, trusted AI technology is also widely used in data privacy protection, financial technology services and other fields. In June this year, the 29th session of the 13th Standing Committee of the National People's Congress passed a number of bills and two decisions, including the data security law. Article 4 of the data security law proposes that to maintain data security, we should adhere to the overall national security concept, establish and improve the data security governance system, and improve the ability to guarantee data security.

This indicates that data security has risen to the level of overall national security concept and linked with national security. All circles pay close attention to this.

"Whether it is AI or science and technology, there are many credible ways, such as stability, interpretability, privacy protection, fairness. How can we combine these to form a unified consensus?" Tao Dacheng, a foreign academician of the European Academy of Sciences, an academician of the Australian Academy of Sciences and a professor at the University of Sydney, has raised his own question. Due to the lack of explainability, it also limits the wider application and empowerment of artificial intelligence. For example, how to determine the responsibility after the unmanned vehicle is in trouble on the road, because the driver can't explain it in a few seconds, Judge the road.

To some extent, due to the lack of standards, the application of artificial intelligence is limited. Tao Dacheng further believes that both the EU's gdpr and China's privacy protection regulations indicate that users should be more cautious about the use of data, indicating that users have considerable rights to protect personal information. Only after the above four capabilities are integrated and clear responsibilities can we find a suitable way to measure trusted artificial intelligence.

There is no doubt that artificial intelligence plays an important role in more fields, such as medical diagnosis, unmanned driving and other scenarios, it is likely that a small algorithm error may lead to fatal disasters. Therefore, the development of certain quantitative indicators and industry standards, in the process of product development, evaluation, can be more stringent evaluation and follow the rules.

The era of intelligence has accelerated, and the most fundamental meaning of the development of science and technology is to improve science and technology“ To be good mainly refers to two aspects. One is whether your intention is good or bad? On the other hand, what are the consequences of doing this. The development of technology has also brought about the problem of fairness. The real improvement of science and technology should be to give everyone more development rights. Everyone can keep up with the times and enjoy the dividend and universal benefits. "

In his opinion, no matter an enterprise or an industry, when it comes to trust, it is necessary to start with responsibility. Only with good intentions can good results be achieved. So as to establish a healthy and orderly development ecology of mutual influence, mutual support and interdependence.