China News Service, Beijing, December 12 (Xinhua) -- Ren Fuji: How does ChatGPT's one-year-old AI "trendsetter" face the "growing pains"?

China News Service reporter Zhang Huandi

A few days ago, ChatGPT, which has been released for one year, ushered in its first birthday. In the past year, ChatGPT, the "trendsetter" of the AI industry, has continued to make progress towards the goal of faster, safer, and cheaper, and has explored the frontier of AI technology development with a group of latecomers. However, AI represented by ChatGPT has also caused many controversies, such as privacy leaks and the change of leadership of ChatGPT's parent company "Open Artificial Intelligence Research Center" (OpenAI) in the United States, etc., which have added a bit of "growing pains" to the development of AI.

Ren Fuji, honorary vice president of the Chinese Society of Artificial Intelligence, chief scientist of artificial intelligence at the University of Electronic Science and Technology of China, and academician of the Japanese Academy of Engineering, said in a recent interview with China News Agency "East-West Question" that ChatGPT has stimulated innovation and competition in the field of AI by science and technology enterprises and research institutions in various countries, and the AI technology represented by it will have a profound impact on all disciplines or industries with potential big data research resources. The various problems caused by the development of AI cannot be ignored, and relevant parties such as practitioners, researchers, and regulatory authorities should take timely measures.

The transcript of the interview is summarized below:

China News Service: It's been a year since ChatGPT was publicly released. What changes and improvements have been made to ChatGPT itself during the year, and what are the shortcomings?

Ren Fuji:ChatGPT's massive progress during the year was largely due to the release of OpenAI's AI language model, GPT-4. Compared with GPT-3.5 and other large language models, GPT-4 is more reliable and creative in handling complex tasks, and can also process more subtle instructions, which tends to be a stable, reliable and credible multi-language multi-functional multimedia intelligent system.

Specifically, GPT-4 has more natural language processing capabilities, more efficient text generation capabilities, stronger creativity and adaptability, and strong self-learning Xi self-evolution capabilities. These features allow it to understand human language more accurately, generate high-quality natural language text based on context and semantics, and process visual content such as images and videos. GPT-4 can also improve performance and evolve based on accumulated data and experience to better adapt to human needs.

A young man accesses and tries out GPT-4. Photo by Wang Gang

However, GPT-4 still has deficiencies in the timeliness of knowledge, the authenticity of responses, and the legitimacy of application, which are manifested in fabricating facts and violating personal privacy.

China News Service: After ChatGPT became popular, technology companies and research institutions around the world have followed suit and launched self-developed competing products. What is the significance of the birth of ChatGPT, and what impact and inspiration does it have on many newcomers?

Ren Fuji: At present, OpenAI's GPT series models are the most representative and leading large models in generative AI, and most of the large language models are structured in GPT-like patterns. The emergence of ChatGPT has undoubtedly had a profound impact on the global technology community, stimulating innovation and competition in the field of AI by technology companies and research institutions in various countries.

ChatGPT shows the "emerging" phenomenon of large language models. When the model size of a language model reaches a certain level, the model will "emerge" with some capabilities that small models do not have, such as few-shot prompts, multi-step reasoning capabilities, instruction execution, program execution, model correction, etc. This has greatly raised people's expectations for artificial general intelligence (AGI), catered to the urgent needs of various groups of people for generative artificial intelligence content (AIGC) in their work, life, and creation, led the technical route of large language models, and strengthened the confidence of a group of latecomers to follow and move forward. AGI refers to an AI system that has no less than human intelligence in any professional field that humans have been or will be involved in.

Referring to the current interaction cases in the general sense, it is not difficult to see that ChatGPT has milestone significance in the underlying data, core technology, application scenarios, user experience and other aspects, and its biggest highlight is its excellent language understanding, intention recognition, logical reasoning, content generation and other capabilities. Its emergence means that the interactive ability of AI has developed by leaps and bounds, to a certain extent, it meets the public's needs for the recognition and use of AI, and makes humans realize that AI may really surpass and replace humans.

China News Service: A few days ago, the storm over OpenAI's change of leadership has attracted widespread attention. Reuters quoted sources as saying that several researchers within OpenAI have developed an extremely advanced model "Q-Star" to seek a breakthrough in the field of AGI, which has sparked controversy and disagreement within the company over AI security, which in turn has triggered the change of leadership. What impact will the incident and possible "Q-Star" have on ChatGPT and the AI industry as a whole?

Ren Fuji: Judging from the current media disclosure, OpenAI's change of leadership involves AI security issues. The essence of this turmoil is the dispute over the route of the AI development concept, that is, the conflict between the concepts of "effective accelerationism" and "super 'love' alignment". "Effective accelerationism" advocates that human beings should unconditionally accelerate technological innovation and subvert the social structure through rapid iteration, and the social risks that may be brought about by technological disruption will essentially force human beings to continuously adapt and reach a higher level of consciousness. The concept of "super 'love' alignment" is to make the AI system do things according to clear instructions, so that AI meets human requirements, and the core is to benefit mankind.

Some scientists at OpenAI believe that the team led by CEO Sam Altman has focused on enhancing AI's capabilities and neglected to enhance the emotion of AI, which is the love of humanity.

If "Q-Star" exists, it could represent a major step forward for OpenAI in the field of AGI and a giant leap forward in the field of AI, as most current AI systems are still experts at specific tasks, rather than generic solutions. True AGI is able to understand, Xi and apply knowledge just like humans.

Many AI experts tend to believe that "Q-Star" exists and can improve the capabilities of existing AI, but it is still far from AGI. But regardless of whether "Q-Star" exists or not, OpenAI is still leading the way in the AI field, and the technologies and models it develops will continue to have a profound impact on the entire industry.

China News Service: Since the beginning of this year, the public has witnessed the progress of AI in areas such as image and audio and video processing, and now AI can not only write articles, but also create paintings, write music and even make short films. How do you see the development trend of AI in various fields?

Ren Fuji: In addition to knowing astronomy and geography, AI can also write written materials, create paintings and music, make short films, and even have a continuous and fluent dialogue with humans with high emotional intelligence. The versatility of AI means that it has great potential to replace human work, and even creative work that is "unique to humans", such as artistic creation and scientific discovery.

Visitors watch the AI "Painting" Dali immersive art exhibition in Shougang Park. Photo by Tian Yuhao

Traditional natural sciences have presented a new model of "AI + natural sciences". By helping to break through the inherent cognitive limitations of human beings, AI represented by ChatGPT is expected to become an important driving force for the advancement of natural science in the future.

At the World Artificial Intelligence Conference, GE's Edison Ecosystem 2.0 expanded the service field of AI from enhanced imaging to improved ultrasound standardization to smart operating room applications. Photo by Chen Yuyu

Taking life sciences as an example, AI technology has made practical progress in the fields of bioinformation retrieval and extraction, small molecule generation, and protein structure research. In the future, AIGC will effectively promote the step-forward development of small molecule generation and optimization technology, small molecule and protein binding conformation prediction, protein and peptide sequence design, and human genomics analysis.

In general, all disciplines or industries with potential big data research resources will be affected by the current AI paradigm of "big data + deep learning Xi". In the field of AI digital content generation, AI has demonstrated its creative potential on the basis of big data Xi, and the new pattern of "AI + cultural industry" has taken shape.

China News Service: The explosion of ChatGPT has once again brought the concept of AI back to the public eye, and countries have also increased their investment in the field of AI. What is the current stage of China's AI industry?

Ren Fuji: In terms of the number of scientific research papers, the scale of investment and financing, the number of enterprises, and the number of patents, in recent years, the development of China's AI industry as a whole has been in an international advanced position, especially in the application of AI scenarios. However, the overall quality of development needs to be improved, especially in the field of core technologies such as hardware and algorithms, and most Chinese companies are still inferior to international AI giants in terms of talent, core technology, and intellectual property rights.

Visitors experience AI calligraphy robots at the first China International Supply Chain Promotion Expo. Photo by Du Jianpo

From the perspective of AI cutting-edge engineering technology, China's technical level in the field of intelligent service robots ranks among the top in the world, especially in the fields of vision, sensing, multi-modal human-computer interaction, affective computing, etc., but there is still a certain gap between mechatronics and robot control and the world's first-class level. However, thanks to the broad market, a wide range of application scenarios and other factors, China has the possibility of surpassing in the field of intelligent service robots.

China News Service: Nowadays, many people on the Internet use AI to collect the voices of celebrities, and after processing them, they can make the AI say anything under the guise of celebrities. Some self-media also use AI to make "photographic pictures" to fabricate news and gain traffic. What problems may arise from the continued advancement of AI technology, and how should AI practitioners, researchers, and regulators respond?

Ren Fuji:In addition to the convenience that AI can bring to human beings as a production tool, it has also raised many issues, such as: unemployment, traditional jobs and even creative jobs that were originally thought to be replaced by AI, privacy protection issues, data abuse threatens users' privacy, security issues, fraud and fabrication generated by AIGC, and ethical issues, including social responsibility, legal ethics, intellectual property rights, emotional attribution, etc.

There are four ways to deal with the various problems that may be caused by AI: improving the governance system of science and technology ethics, strengthening the guarantee of science and technology ethics governance system, strengthening the review and supervision of science and technology ethics, and carrying out in-depth education and publicity on science and technology ethics.

While following the principles of science and technology ethics, it is also necessary to ensure the dynamic adjustment of science and technology ethics governance, not to overemphasize the possible negative impact of science and technology development, to scientifically respond to unforeseen problems, and to ensure the freedom and vitality of scientific research.

China News Service: What does your ideal perfect AI look like?

Ren Fuji: My ideal perfect AI is a combination of AGI (creator) and dedicated artificial intelligence (domain expert), pursuing the unity of ability and emotion, and the unity of technology and ethics.

China's first artificial intelligence-themed dance drama "Deep AI You" was staged at the National Centre for the Performing Arts in Beijing. Photo by Jia Tianyong

The so-called "creator" does not refer to the creator in theism, but is an analogy of AGI in terms of ability and potential. AGI must not only lead humans in the field that it is good at, but also surpass it in the field that it is not good at. The more advanced AIGC formed on this basis will demonstrate unprecedented methods and unexpected effects in the fields of problem solving, artistic creation, and content production. This is the role that AGI exhibits is similar to that of the Creator.

AGI is the original intention and ultimate goal of AI research, but it will not be achieved in the short term. At present, we refer to AI systems that focus on subdivisions and solve real problems as specialized artificial intelligence. At present, the relatively mature application of industrial visual inspection, production line assembly control, etc. are all application examples of dedicated artificial intelligence.

In addition, considering the security threats that AI may bring, we should also deploy and apply AI systems in specific fields first. With the improvement of AI interpretability and the confirmation of the safety and controllability of AI systems, the application space of AI systems will be gradually expanded. (ENDS)

Interviewee Profile:

Ren Fuji, Academician of the Japanese Academy of Engineering, Doctor of Engineering from National Hokkaido University, Chief Scientist and Distinguished Chair Professor of Artificial Intelligence at the University of Electronic Science and Technology of China, Chief Scientist of the Institute of Artificial Intelligence, Chief Scientist of the Shenzhen Institute for Advanced Study, President of the International Institute of Advanced Informatization, and Honorary Professor of Tokushima University. He has served as a CSK Fellow in Japan, a visiting professor at New Mexico State University, a visiting professor at Florida International University, a visiting professor at Harvard University, and the dean of the Department of Intelligent Engineering, graduate dean, and director of the Information Decision Department at Tokushima University in Japan.

He is a National Distinguished Expert, Chair Professor of the Changjiang Scholars Award Program, Recipient of the Overseas Outstanding Young Scholars Fund, Honorary Vice Chairman of the Chinese Society of Artificial Intelligence, Haizhi Expert of the China Association for Science and Technology, and Member of the Scientific and Technological Expert Advisory Committee of the Overseas Chinese Affairs Office of the State Council. Founder and chair of the IEEE International Conference on Natural Language Processing and Knowledge Engineering. The first president of the Chinese Science and Technology Scholars Alliance in Japan, the former president of the New Overseas Chinese Association of Japan, and the former president and honorary president of the All Japan Chinese Doctoral Association. He is the winner of the Japan Natural Science Endo Award, the Recreation Club Award, the first prize of the Wu Wenjun Artificial Intelligence Science and Technology Award for Innovation, the "Outstanding Contribution Award" of the Scientific Chinese Person of the Year, the Huangshan Friendship Award of the Anhui Provincial Government, and the Friendship Award of the Chinese Government.