In the field of artificial intelligence, we must believe that "good students make good teachers"
2024-08-17
한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina
Hong Kong University of Science and Technology Chief Vice-President Guo Yike (second from right) takes a photo with a Global Times reporter after an interview.
The "Greater Bay Area of the Future" drawn by artificial intelligence developed by the "Hong Kong Generative Artificial Intelligence (AI) R&D Center" shows the combined appearance of the 11 cities in the Guangdong-Hong Kong-Macao Greater Bay Area imagined by the machine.
Our special correspondents to Hong Kong are Yang Shasha, Zhang Xueting and Zhang WeilanAt this year's "Hong Kong International Innovation and Technology Expo 2024", technologies such as "Artificial Intelligence (AI) turns stories into videos in seconds", "Artificial Intelligence draws the "Greater Bay Area of the Future Thousand Miles"", and "Encounters across time and space" allow the outside world to see the latest research and development results of Hong Kong in the field of artificial intelligence.The above applications are all made through models independently developed by the "Hong Kong Generative Artificial Intelligence Research and Development Center" (HKGAI), the director of which is Professor Guo Yike, the Chief Vice-President of the Hong Kong University of Science and Technology. Public reports show that Guo Yike's family has three generations of Tsinghua people. In 1980, Guo Yike was admitted to the Computer Science Department of Tsinghua University and became a well-known scientist in the computer field. At the same time, he also engaged in artificial intelligence research very early and is an internationally renowned scholar in the field of artificial intelligence. Four years ago, Guo Yike came to Hong Kong and served as Vice-President of Hong Kong Baptist University and Chief Vice-President of the Hong Kong University of Science and Technology.In his office at the Hong Kong University of Science and Technology, the Global Times reporter's exclusive interview with Guo Yike started with the "Hong Kong version of the big model" and talked about how Hong Kong and the mainland are innovating in the field of artificial intelligence. He particularly emphasized the need to believe that "good teachers make good students" in the cultivation of artificial intelligence talents, and believed that artificial intelligence is the world of young people, "a lot of things were taught to me by my students."“Nothing is right”Global Times: According to public information, HKGAI's first self-trained basic model has been basically completed. It is the first basic model independently developed by Hong Kong in cooperation with the mainland. Hong Kong officials have recently begun to try out this "Hong Kong version of ChatGPT" as a "civil servant document assistance system." In addition, HKGAI has also tailored vertical applications based on this basic model for different fields such as law, medical care and creativity. Please give us examples of the specific applications of these big models.Guo Yike: The Hong Kong government writes official documents very frequently, with its own fixed format, mainly in English. The "Civil Servant Document Assistance System" can help find information and generate content. Civil servants can also organize their own views on the generated content. We are currently trying it out in some Hong Kong government agencies and are also training the data model based on feedback. If the trial process goes smoothly, it will be gradually expanded to other government departments and will eventually be open to the entire Hong Kong society. I believe similar applications will also be promoted in the mainland, and the mainland will also do well.Taking law as an example, the current application of artificial intelligence at the legal level is mainly legal information query. In the future, we can extend it to case analysis. For example, tell the big model about the dispute between the parties, and then the big model will tell the parties the legal basis, suggest litigation methods, and even analyze possible results based on existing cases. In the future, it can also be extended to conversational analysis, like a real lawyer, talking to the parties.Of course, this will also bring some challenges to lawyers, and the way lawyers work will change greatly. The personal value of lawyers must be raised higher, and they must do things that machines cannot do. "When the car came out, the coachman disappeared, and the driver was born." This does not mean that lawyers will be eliminated in large numbers in the future, but as the way of working changes, new professions will emerge in society.Global Times: Baidu founder Robin Li recently stated at the 2024 World Artificial Intelligence Conference that the development path of AI technology has undergone a directional change, calling on everyone to "not focus on models, but on applications!" What do you think of this?Guo Yike: I think that in the field of AI, the so-called volume is repetition, meaningless and unprogressive repetition. I think it is wrong to be volumetric, and it is wrong to be volumetric in applications, and there is nothing to be volumetric about big models. It is boring for everyone to repeat the same mature technology, and it will also lead to a waste of resources. "The market only needs a few big models" or "After a few leading companies or government agencies create a universal big model, no one needs to do it anymore" are worth discussing. The ideal state, of course, is to have a relatively stable, efficient, and mature basic big model, but I do not agree that this basic model does not need to be studied or constructed.I think that large models need to be made deeper, and to achieve general artificial intelligence, there must be a "foundation". When we look back three years later, the "foundation" at that time will definitely be much stronger than the current one. However, we need to build a brand new "foundation" and solve many existing problems. Only after these problems are solved can we update it generation by generation. This is very necessary.The US startup OpenAI is also constantly iterating, and every time there is progress. Every iteration must be innovative and solve problems. Otherwise, we will always be catching up and will never make progress. If there is progress, iteration is necessary, but each iteration must be driven by innovation. In addition, don't worry about what others do, don't worry about what the United States or other countries do. The key issue is that we must spend time on technology, find problems, study them carefully, move forward step by step, and have a clear direction.“Artificial intelligence is the world of young people”Global Times: What are the advantages of developing artificial intelligence technology in Hong Kong?Guo Yike: Hong Kong's biggest advantage is talent. It can gather talent in a short period of time and can also communicate extensively with the international community. There are few barriers. Hong Kong's innovation ecosystem is better and more autonomous. It is not too trendy or too competitive. Hong Kong's innovation soil and ecosystem are suitable for the growth of talents. If you follow the trend, you will never be able to innovate. You will just be struggling in someone else's system. For Hong Kong, it is still very important to maintain an innovative mindset and spirit.There are several milestones in the development of artificial intelligence in China, and Hong Kong has a shadow in each of them. SenseTime is the best example. Tang Xiaoou of the Chinese University of Hong Kong has been engaged in computer vision research for a long time, and founded SenseTime Technology to make it the mainstream in the industry. In addition, DJI drones, founded by Wang Tao of the Hong Kong University of Science and Technology, are actually artificial intelligence products. Flight, control, photography, etc. are all standard artificial intelligence technologies. Now the market summarizes these technologies as "embodied intelligence." But when Wang Tao was doing research and development, he may never have thought about this concept. He just wanted to solve the problem and make the camera fly.Global Times: The Hong Kong University of Science and Technology will launch an extended major in artificial intelligence in 2020. It is said that students can study artificial intelligence technology in addition to their major. What factors does HKUST pay most attention to when cultivating talents?Guo Yike: In the 2024 QS World University Subject Rankings, the Hong Kong University of Science and Technology's "Data Science and Artificial Intelligence" subject ranked tenth in the world. The QS ranking places more emphasis on the training of students, which is also a very good affirmation of our education and training in the field of artificial intelligence.We have always advocated that anyone can pursue a degree in artificial intelligence in any major. For us, artificial intelligence is a universal and practical technology. We did not put artificial intelligence in the computer department, but covered it throughout the school, biology plus artificial intelligence, chemistry plus artificial intelligence, machinery plus artificial intelligence, etc. We have also spent a lot of energy on several major directions of artificial intelligence research, and our artificial intelligence computing resources are among the best in Hong Kong universities and the Greater Bay Area.We have always emphasized a point of view that in the emerging field of artificial intelligence, it is not "good teachers produce good students", but "good students produce good teachers". Our chairman of the board of directors, Harry Shum (foreign academician of the U.S. National Academy of Engineering, foreign academician of the British Royal Academy of Engineering, and world-class expert in computer vision and graphics research - Editor's note) has always told me that the field of artificial intelligence is a place where "good students produce good teachers". Artificial intelligence is the world of young people, and many things are taught to me by students. Eating and chatting with students is a very good way, because they have strong judgment and are exposed to the latest research. The advantage of us old guys is that we have experience. You tell me some fragmented information, and I can put it together into a better theory and tell you what the relevant research was like many years ago.A good university, a world-class university, must have equal academic exchanges and cooperation with students. This is a basic academic environment. Artificial intelligence is developing too fast. Unlike traditional Chinese medicine, where older practitioners gain more experience, artificial intelligence has new methods and new ideas every day. You must work with young people to keep up with the trend.No need to compare the gap between China and the United StatesGlobal Times: Among computing power, data and algorithms, which one is the most difficult for China right now?Guo Yike: We should still spend more energy on algorithms and research, and not blame everything on lack of computing power. We are definitely not the country with the strongest computing power in the world, but we can be regarded as the second strongest country in computing power. In the field of artificial intelligence research, we are not as good as the best, but better than the worst. Moreover, we have very strong resource scheduling and adaptation capabilities.I think the main problem in China's AI field is innovation, innovation in basic science. A recent report released by the World Intellectual Property Organization shows that from 2014 to 2023, China's generative AI patent applications exceeded 38,000, ranking first in the world, six times that of the second-placed United States. This is a patent, and there is no direct and necessary connection with innovation.Algorithms are actually directly related to our country's innovation capabilities. Whether in Hong Kong or the mainland, we still have a certain gap in innovation capabilities with Silicon Valley in the United States. The ability, environment, and ecology of innovation are what we really need to catch up with, and this is the most difficult. Students who have very good test scores in college are often not the final innovators. Wang Tao's grades were not the most outstanding in school, but later he founded the unique DJI with the support of the Hong Kong University of Science and Technology. The school should be extremely tolerant of students' academic research and encourage students to solve problems they think are useful to mankind according to their own hobbies and thinking.Global Times: Last month, the US company OpenAI officially banned users in some countries from accessing its services. The latest restrictions by OpenAI have caused some people to worry about "not being able to use the world's leading large-model products." What do you think of the impact of OpenAI's "supply cut"?Guo Yike: To be honest, I don’t care too much about this. OpenAI may give us some inspiration. But we can’t take shortcuts in technical research. We still need to follow our own path, learn more solidly, and not be influenced by the outside world. Whether we are blocked or not, the key is that we must have our own ideas. Sometimes the blockade and restrictions of the outside world may make your own ideas freer. If others don’t open source, it may not be a bad thing for us to do it ourselves.Take DJI as an example. Why did DJI do so well? Because it had no predecessors to learn from, and it has come forward step by step and has now become one of the best drone companies in the world.In the field of artificial intelligence, we don't need to compare the gap between China and the United States. Artificial intelligence is not a field of linear development. It covers a wide range and is blooming in many fields. What we need to think about is: what do ordinary people, the whole society, the country and even the entire human race need? We should make these things and do them early and well. What will the big model look like in the next 5 or 10 years? How to make up for the defects of the big model today? What are the fundamental problems? How can we innovate? Thinking about these is much more interesting. ▲
