Generative artificial intelligence has recently ushered in "rocket" development, and it has naturally become the hottest topic that cannot be avoided at this World Artificial Intelligence Conference. Guo Yike, chief and vice president of the Hong Kong University of Science and Technology, joked that the entire conference seemed to have become a "world ChatGPT conference". While generative artificial intelligence greatly liberates productivity, how to regulate safety, legality, morality, ethics and other issues?
On July 7, the 2023 World Artificial Intelligence Conference Intelligent Society Forum was held in Yangpu Riverside. With the theme of "Intelligent Society and Digital Governance Ecology", Chinese and foreign experts and scholars participated in the forum to conduct in-depth discussions on these issues, and released a number of special research results such as "Guidelines for the Ethical Norms of Artificial Intelligence Large Models" and "Generative Artificial Intelligence Risk Assessment Framework (1.0)".
Balance AI development and security
The benefits of artificial intelligence technology need not be repeated, and the experts at the meeting basically agreed: it can accelerate the progress of scientific and technological innovation, stimulate productivity and economic growth... But experts also agree that the risks of AI technology cannot be ignored.
In her video message, Gabriella Ramos, Assistant Director-General of UNESCO's Social and Human Sciences Sector, mentioned that AI technology will replicate and amplify discrimination, thereby deepening social inequality; The environmental impact cannot be ignored, with the CO2 emissions generated by the large model equivalent to 125 flights to and from Beijing and New York; Large language models also generate a lot of errors and false information.
The key is to balance development and security. Su Jun, Dean of the Institute of Intelligent Social Governance at Tsinghua University, gave an example from the perspective of AI and environmental protection: "At present, there is still a corner between digitalization and green development, and if handled properly, intelligent technology will play a huge role in promoting the revolution of energy production and consumption, greatly reducing energy consumption and carbon emissions, and helping to achieve the 'dual carbon' goal as soon as possible." Conversely, improper handling may lead to more energy consumption and loss of social welfare, plunging society into an ecological crisis. ”
"In order to effectively balance development and security, exploring the governance framework of artificial intelligence has become the consensus of the international community." Ye Linlin, deputy director of the Municipal Education Commission, said that the EU's AI legislation has made great progress, and the Cyberspace Administration of China has also drafted a draft of the management measures for generative AI services.
The "Operational Guidelines for Ethical Norms for Artificial Intelligence Large Models" (hereinafter referred to as the "Operational Guidelines") released this time is aimed at addressing the risks of large models of artificial intelligence and promoting the implementation of ethical norms. Du Yanyong, the relevant leader of the research team and distinguished professor of the School of Humanities of Tongji University, said that most of the ethical principles and guidelines of artificial intelligence are convergent and overlapping, indicating that a high degree of consensus has basically been reached on the core ethical principles of artificial intelligence. However, how to implement these ethical principles, but there has never been a strong measure, which is also the fundamental reason for trying to explore the operational guidelines of the ethical norms of artificial intelligence large models, "We point to the technical characteristics and possible application scenarios of artificial intelligence large models as much as possible to improve the pertinence of the operation guidelines." ”
"Humans" must be in the decision-making loop
According to reports, the operation guidelines focus on the research and development of artificial intelligence. The research team believes that researchers bear the most important moral responsibility in the ethics of artificial intelligence. Notably, the guidelines emphasize "human" supervision, control, and responsibility.
"We believe that to a certain extent, the decision-making of AI big models can be aligned with human interests and values, which is also the most important aspect of realizing the safety of AI big models." Du Yanyong said, "No matter how strong the ethics of the artificial intelligence big model, after all, it still has tool attributes. Therefore, we must ensure that when it comes to major decisions, 'humans' must be in the decision-making loop. AI big models must be under human supervision and control. ”
To this end, the operational guidelines propose that by improving the distribution of responsibility content, a closed loop of responsibility should be formed, so as to avoid the so-called responsibility gap and organized irresponsibility as much as possible.
Guo Yike also believes that the most important problem in AI risk management is that when AI "replicates" human performance through incredible technical means, it may disrupt human cognition, making humans mistakenly think that chatbots are "thinking in a human way", which will lead to a series of consequences.
In fact, artificial intelligence only behaves like a human, and the "thinking" process is more like a "black box". Guo Yike told the story of the movie "Her": the male protagonist and his wife were very upset after divorce and began to chat with a computer similar to ChatGPT. The computer-simulated female voice is very good, and she is gentle and humorous. The male protagonist falls in love with AI out of control, only to find out in the end that AI will be routinely gentle and humorous to everyone. If the "she" of AI can be regarded as falling in love with the male protagonist, then "she" is also in love with many people at the same time. This difference in perception ultimately destroys the psychology of the male protagonist.
"This film makes an accurate prediction of today's situation." Guo Yike said, "When we govern AI, we must first let people understand machines, let machines understand people, and let machines and people empathize, so that such governance will have a solid scientific foundation." ”
To manage, but also dare to try and dare to break in
However, experts such as Du Yanyong also stressed that the governance (or management) of AI is not to hinder or prohibit the development of AI technology, "which no one can do." ”
Shen Tiyan, a researcher at the Institute of Artificial Intelligence of Peking University, said that AI governance is mainly three problems, creating AI well, using AI well, and managing AI well, "In the latter two aspects, China has absolute advantages." But in terms of original innovation, we still have some shortcomings. He suggested that AI developers should be given a little space and be more tolerant of them, "try to pay less attention to them, and toss them around." Toss out the results, and then find a way to manage. ”
"We can't adopt an ostrich policy, let alone a blocking policy, don't worry about uncontrollable technology, choose not to use or use less." Su Jun said, "The door of science and technology has been opened, and we should bravely stand on the stage, take active governance measures to embrace new technologies, and comprehensively strengthen the application based on ChatGPT architecture." ”
He mentioned that Yangpu District is one of the first ten national intelligent society governance experimental comprehensive bases in China, and it should take the lead and continue to explore, constantly enter the no-man's land, and constantly enter the application scenarios of new AI technologies that have never been seen before, promote more technology applications, and make new technologies better serve high-quality development. Shan Bin, director of the Information Coordination Division of the Cyberspace Administration of the Municipal Party Committee, also believes that Yangpu undertakes the mission of grasping pilots, breaking new roads and gaining experience for intelligent governance in Shanghai and even the whole country.
The reporter learned from Yangpu District that Yangpu is closely combining urban renewal and digital transformation to promote scene construction and experimental research, and initially formed a social experiment model of "answering questions" by application subjects and "answering questions" by research subjects and technical subjects. Since the beginning of this year, according to the relevant requirements of the Central Cyberspace Administration of China, Yangpu has focused on the opportunities and challenges brought by generative artificial intelligence big model technologies such as Chatgpt, and cooperated with Bilibili and other enterprises to carry out social experimental research. In addition to the preliminary research results of the "Generative AI Risk Assessment Framework (1.0)" released at the meeting, Yangpu also unveiled the "AI Compliance Service Center". The center will give full play to the intellectual and technical resource advantages of Tongji University and the practical advantages of Yangpu District to provide compliance guidance and consulting services for enterprises' new artificial intelligence technologies and applications.
Ni Bing, executive deputy mayor of Yangpu District, said that Yangpu will deeply analyze the hot spots and pain points in the field of regional economic and social life, constantly raise questions, carry out practice, and build a number of scenes that enterprises and the masses really feel effective, easy to use and love to use. Create a number of demonstration projects of "no one has me, everyone has me excellent", take the lead in forming a number of display windows with obvious advantages, and timely translate research results into policies, standards, norms and measures, and convert more articles into documents, providing more Shanghai demonstrations and national templates for the governance of intelligent society.
Editor-in-chief: Tang Ye Source: Zhu Jun Design
Source: Author: Hu Xingyang