Three questions about the future picture of artificial intelligence——Observation of the thematic forum of the Chinese Gong Intelligence Conference

On July 22~23, the Chinese Intelligence Conference was held in Fuzhou. Fifteen thematic forums took turns, academicians, experts and entrepreneurs, focusing on the theme of "Digital Intelligence Leading the Way and Building the Future Together", focusing on the research progress and academic frontiers of artificial intelligence at home and abroad, providing new inspiration and ideas for the breakthrough of basic theory of artificial intelligence, key technological innovation and industrial application, and drawing a future picture with the wisdom of collision.

One ask

What changes has AI brought about?

With the continuous advancement of science and technology, the application of artificial intelligence is becoming more and more extensive, and its influence has far exceeded our imagination and led change in many fields.

In the new wave of global artificial intelligence technology development, large models are a key word. ChatGPT and others "learn" massive data through large models and make breakthroughs.

"ChatGPT has deep knowledge and strong analytical conversational capabilities, but its 'success' is more proof that generative language models can be a viable path to 'achieve universal AI', and it is far from successful." At the forum on large models and natural language processing, Zong Chengqing, a researcher at the Institute of Automation of the Chinese Academy of Sciences, took machine translation as an example to analyze the capabilities of ChatGPT. He believes that large language models will not become the end of natural language processing, and large language models with both advantages and disadvantages have pushed NLP (natural language processing) and even the entire AI field to a new historical stage and platform period through their research paradigm.

Artificial intelligence large models are undergoing dramatic changes at the application level. At the thematic forum on embodied intelligence and collaborative interaction, Huang Yan, associate researcher of the Institute of Automation, Chinese Academy of Sciences, shared the possible evolution trend of visual-language navigation research under the new background of the rise of large models. Huang Yan introduced that visual-language navigation is an emerging research direction in recent years, mainly operating in the development of humanoid robots, involving very multidisciplinary and technical interaction, although the current performance in the simulation environment is good, but in the process of moving towards the real environment, accuracy and performance will face a 'generation gap'. "Nowadays, with the application of large language models, robots can deduce instructions by mining knowledge in language models, build maps based on visual information, make corresponding symbolic reasoning patterns, make possible decisions, and become more 'intelligent'."

At the same time, the emergence of large models provides a new path and way for the study of graphics and three-dimensional. At the AIGC and Art Innovation Forum, Xin Tong, Principal Researcher of the Network Graphics Group at Microsoft Research Asia, introduced the opportunities of intelligent 3D content generation. "There are some problems with traditional 3D content generation, such as the difficulty of using 3D modeling software and the expensive capture equipment; Production assembly line factory, there are many tools, the use is cumbersome; Requires a lot of manpower and professional skills and training. Over the past few years, the way intelligent media content is generated has led to breakthroughs in image generation. He said that intelligent three-dimensional content generation can bring a new content production paradigm, through text description and sketch control, unified reconstruction and generation algorithms, and the use of pictures to generate large models.

Second question

What are the new applications of AI?

In the thematic forum, the implementation of multiple applications of artificial intelligence is a common concern. With the rapid development of large models as the first march, a new round of scientific and technological revolution and industrial transformation is accelerating evolution, bringing about the closer integration of artificial intelligence technology and industrial development.

At the Robot Perception and Self-Learning Forum, Liu Yong, director and professor of the Center for Intelligent Driving and Future Transportation of the School of Control of Zhejiang University, introduced that in recent years, with the breakthrough of sensor technology and SLAM theory, autonomous mobile robots have moved from research to market application, involving fields including unmanned driving, smart cities, leg-and-foot robots, Mars rovers and so on. International giants such as Google, Apple, and Meta focus on this technology, and domestic giants such as Huawei, Baidu, Tencent, Ali, etc., have also carried out special research. He proposed, "If you look at the autonomous mobile robot, you will find that perception, planning and control are the three core links, of which everyone has done more in terms of control, although the perception and planning technology has made breakthroughs, but there are still some challenges to overcome." ”

The application of artificial intelligence in the medical field is also the focus of many experts and scholars.

At the AIGC and Intelligent Medicine Forum, Zhan Yiqiang, President of R&D and Chief Operating Officer of United Imaging Intelligent Technology Co., Ltd., believed that the improvement of algorithm performance can change the way AI applications are used, and through software architecture innovation, AI technology can be seamlessly embedded in doctors' workflows, forming solutions around multiple AI applications of diseases or locations. Zhan Yiqiang introduced that in a recent case of chronic total obstruction of coronary arteries, artificial intelligence coronary reconstruction software shortened the reconstruction and measurement time by 73%, and achieved an increase in the success rate of reconstruction from 48% to 95% of traditional workstations. "Now, we are trying to use GhatGPT's large language model for large-scale application of text, transform unstructured surgical process records into structured surgical induction, and empower various medical scenarios through the combination of text large models and image models to build more accurate 'professional' large models with medical expertise."

Artificial intelligence not only appears in the diagnosis and treatment stage, but also in the rehabilitation stage.

At the multi-mode perception interaction forum, Li Zhijun, deputy dean and professor of the School of Information Science and Technology of the University of Science and Technology of China, shared a variety of neuroprosthetic limbs such as the wearable robots developed by his team, including brain-controlled wheelchair systems, gait training rehabilitation robots, two-arm exoskeleton robots, paraplegic assisted exoskeletons, bionic robotic arms, and upper and lower limb exoskeleton robot systems. "Our rehabilitation robot control technology is original at home and abroad, not only the fastest walking lower limb rehabilitation robot in China, but also through the integration of brain-controlled active rehabilitation technology, functional electrical stimulation technology, virtual reality technology, etc., to help patients achieve active rehabilitation training." At present, hundreds of clinical trials in Guangdong Industrial Injury Rehabilitation Hospital, Zhongshan Affiliated Medical and other hospitals have been shown in rehabilitation medical application after clinical trials, and the effect is good. Li Zhijun said.

Three questions

Where is AI headed in the future?

With the emergence of a new generation of artificial intelligence technology in the world, injecting new momentum into the development of the digital economy, artificial intelligence is profoundly changing social production and lifestyle. At present, a new round of scientific and technological revolution and industrial transformation has dawned, and in this competition related to the future and destiny of all countries, teams seize opportunities, catch up and strive to surpass.

At the multi-mode perception interaction forum, Professor Zhang Haitao, deputy dean and professor of the School of Artificial Intelligence and Automation of Huazhong University of Science and Technology, introduced the theoretical technology and application of autonomous unmanned vehicle cluster collaborative round-up that he and his team researched. He said that the three major trends in the future development of unmanned systems are autonomy, intelligence and clustering. "In the past, we did formation, later we did swarming, cross-domain clustering, and in the future we want to do wide area, including space, surface, underwater, as well as unmanned systems and fixed base stations, to realize the marine unmanned system community." Zhang Haitao said.

In the broader manufacturing industry, artificial intelligence is also continuing to explore, guiding "Made in China" to "Made in China".

"In order to achieve the goal of becoming a manufacturing power, the mainland has begun to promote the landing of a new generation of artificial intelligence theory and technology in the manufacturing industry, forming a new model, new means and new format of intelligent manufacturing." At the industrial intelligent robot forum, Professor Ren Lei of the School of Automation/School of Software of Beihang University threw out the concept of "artificial intelligence 2.0", believing that what kind of needs and industrial application scenarios the new generation of artificial intelligence can meet in industrial manufacturing is a problem that researchers in the field of automation should think about. Ren Lei said that the new generation of intelligent manufacturing in the future should realize the efficient division of labor and mutual assistance of human-information-physical systems, from "teaching them to fish" to "teaching them to fish", organically combining human wisdom, experience and the powerful learning ability of AI and promoting each other.

At the same time, how to actively govern and effectively govern in the wave of technological change and develop responsible artificial intelligence has become a common issue of the times and an urgent issue for the world. Wang Yingchun, deputy director and researcher of the Governance Research Center of Shanghai Artificial Intelligence Laboratory, led the team to make an attempt. Wang Yingchun believes that the credibility evaluation of large models itself is very complex, and different countries, regions and organizations have different standards, which makes it difficult to reach a complete consensus. It involves multiple levels and superimposed various scenarios, and it is difficult to evaluate with unified tasks and indicators; Compared with aptitude assessments, credible assessments have fewer datasets and tools, and rely more on manual labor. Therefore, he believes that evaluation should become an important discipline to promote the development of LLMs and other AI models. At present, they have initially built a dandelion AI governance open platform, explored the construction of a governance innovation system that integrates "rules-technology-scenarios-evaluation", built a systematic and practical AI ethics and governance infrastructure, implemented the principles of service governance, and supported the improvement of AI governance effectiveness. (Reporters: Jiang Yachen, Yan Xiao, Wu Huazhen, Song Yimin)

Source: Fuzhou Daily