It is the most interesting direction of large digital models. How to make machines have values and common sense.
The sun is high, the mountains and rivers are flowing, and traditional landscape painting techniques are integrated with the unique lines of stone and wood carvings, presenting a new perspective. At the 79th Cultural Forum of Liberation Daily held yesterday, this artificial intelligence art work titled "Carving Mountains and Rivers" was breathtaking. This painting is like a microcosm, showing people how the wings of technology and culture can inherit history, empower innovation, and light up the future.
At this forum, guided by the Propaganda Department of the Shanghai Municipal Committee of the Communist Party of China and the Shanghai Science and Technology Commission, sponsored by Liberation Daily and co-sponsored by Shanghai Caohejing Emerging Technology Development Zone Development Corporation, five people were deeply involved in innovation, resonated with each other, Leading Chinese and foreign representatives discussed the theme of "innovative ecology and new productivity".
Michael Levitt, winner of the 2013 Nobel Prize in Chemistry and vice president of the Association of Top Scientists in the World. Pictures taken by Haishar
"My life changed in October 2022, when ChatGPT 3.5 was released." Michael Levitt said frankly that he uses applications developed based on this large model at least 20 times a day.
Artificial intelligence is undoubtedly a hot topic in science and technology at the moment. However, the 2013 Nobel Prize winner in Chemistry is more willing to explain his understanding of artificial intelligence from an evolutionary perspective.
"The evolution of living things is very interesting." He said that at the beginning, the form of life was very simple. As time went on, eukaryotes began to appear, becoming more and more complex and therefore more diverse. Artificial intelligence also learns from evolution and continuously expands the possibility of information technology serving human beings.
Yike Kuo, academician of the Royal Academy of Engineering and Chief Vice-President of the Hong Kong University of Science and Technology, then Michael Levitt raised his own thoughts: If large models are the answer, what is the problem?
Academician of the Royal Academy of Engineering and Chief Vice-President of the Hong Kong University of Science and Technology, Guo Yike.
In his view, the number of large artificial intelligence models has increased exponentially after the emergence of ChatGPT. Why are artificial intelligence dialogue systems so important? In his view, language is not only a communication tool, but also the encoding of thoughts, which determines the way of thinking. If machines learn human language, they may be able to learn human thinking.
What is the next development trend of large models? Guo Yike believes that it is becoming more and more like a human being, not only with human thoughts, but also with human thinking and expression. If we say that the growth and learning process of a person is from the curious acceptance of common sense in childhood, to the shaping of values in adolescence, from systematic learning of indirect knowledge from textbooks, to self-learning from social experience as an adult; then artificial intelligence is the other way around. : At the beginning, scientists fed all the data to the computer, trained a large model, and made vertical fine-tuning. Then they found that it lacked values, so they started to learn values. The most difficult thing is to abstract all data and phenomena into common sense. Therefore, how to make machines have values and common sense is the most interesting direction of large models.
"In the future, artificial intelligence's answers will not only be words, but also actions. For example, if you see that the tea cup is about to fall, the machine will know how to catch it." Guo Yike said, "Computer science will develop towards machine behavior. , this is a development direction.”
Data, algorithms and computing power: the infrastructure of the future
"For Shanghai, large models should be infrastructure like subways and high-speed rails." Guo Yike believes that an industrial ecology should be formed based on large models, and various high-level vertical applications should be derived, such as medicine, art, finance, and Governance etc.
Also emphasizing infrastructure is Yang Fan, co-founder of SenseTime and president of the large device business group.
Yang Fan, co-founder of SenseTime and president of the large device business group.
Yang Fan focused on a big challenge facing the industry - the artificial intelligence industry's relatively low profit margin. When can artificial intelligence companies make money? When can we sustainably make profits? He believes that with the development of the industry, the continued agglomeration of talents and the expansion of scale groups, the industry will be refined and differentiated - a few people will build the infrastructure of AI and provide low-cost and low-cost secondary creation and applications of artificial intelligence. Threshold and high-efficiency services; others can build applications around this infrastructure, so the technical threshold is much lower than before.
Yang Fan believes that the three elements of artificial intelligence—data, algorithms, and computing power—will be opened up and provided in the form of infrastructure in the future to be used by a prosperous and open downstream industrial ecology. "This is why we feel that artificial intelligence is the core representative of new productivity." He said.
"We are talking about innovation today, but don't forget that innovation is not just technological innovation. The prerequisite for technological innovation is institutional innovation first," said Wang Zhan, chairman of the Shanghai Federation of Social Sciences.
Wang Zhan, Chairman of Shanghai Federation of Social Sciences.
He thought about innovation from the perspective of human history: "Innovation actually has two stages. The first stage is the innovation of the agricultural society, and the focus of this innovation is mainly in China. The second stage is the innovation of the industrial revolution. First, occurred in the West. "The innovation of agricultural society is reflected in the handicraft industry, "such as silk, tea, and porcelain in Jiangnan, which bring together the essence of ancient Chinese handicraft industry." In comparison, the innovations of the Industrial Revolution were much more powerful, disruptive, and influential than the innovations of the agricultural society.
In Wang Zhan's mind, Shanghai-style culture is the international Jiangnan culture, the innovative result of Jiangnan culture absorbing foreign advanced industrial civilization; while red culture is the result of the innovation of Shanghai-style cultural system. Today, the profound accumulation of the three cultures provides rich cultural soil for our technological innovation. In the era of artificial intelligence, technological innovation also provides new means and tools for cultural and artistic creation.
Gao Bowen, vice chairman of the Shanghai Quyi Artists Association and head of the Shanghai Pingtan Troupe, brought his own thoughts and practices on how traditional culture and art can pay attention to and leverage technological development and innovation in the digital age.
Gao Bowen, Vice Chairman of Shanghai Quyi Artists Association and Head of Shanghai Pingtan Troupe.
In April this year, the Shanghai Pingtan Troupe completed an experimental performance with the support of many parties: Mange Xing - AI music narrative across time and space. In his view, as an art creator, you must understand, experiment, and explore artificial intelligence and other new technologies. "You need to have the courage: Can I subvert a work?" Gao Bowen emphasized, "The premise is that you must have enough respect for traditional culture and have a profound foundation in traditional culture before you can innovate."
