From April 25 to 29, 2024, the 2024 Zhongguancun Forum with the theme of "Innovation: Building a Better World" was held in Beijing. As one of the parallel forums of the Zhongguancun Forum, the General Artificial Intelligence Forum was held on April 27. The forum was hosted by the Beijing Municipal Science and Technology Commission, the Zhongguancun Science and Technology Park Management Committee, and the People's Government of Haidian District, Beijing, and organized by the Beijing Institute of General Artificial Intelligence (hereinafter referred to as the "General Research Institute"), the Institute of Artificial Intelligence of Peking University, the School of Intelligence of Peking University, and the Department of Automation of Tsinghua University.
Figure: General Artificial Intelligence Forum
Four major scientific research achievements were released, leading the course of general artificial intelligence with original innovation
General artificial intelligence is the commanding heights of global scientific and technological competition, and it is also a typical representative of new quality productivity. Building general-purpose agents that can understand the real world is a necessary condition for moving towards general AI. To this end, the general artificial intelligence large-task simulation platform "TongSIM" (TongSIM) developed by the Institute provides a "home" with a high degree of simulation with the real world for the general agent "Tongtong". TongSIM has functional features such as virtual environment construction, physical and causal logic simulation, and agent environment perception, which can provide a rich task training environment for general agents and support comprehensive testing of the intelligence capabilities of general agents. In the future, TongSIM will not only continue to accompany the growth of "Tongtong", but also explore the application space in the expanded fields such as embodied intelligence and robot services.
Figure: General Artificial Intelligence Large-task Simulation Platform "Tongjie" (TongSIM)
As the world's first general-purpose agent, the emergence of "Tongtong" has become the biggest highlight of this achievement release. Based on the original theoretical framework of cognitive architecture and the self-developed learning and reasoning framework, "Tongtong" has a unique "mind", which is different from data-driven AI, and is driven by value and causality, which can make explainable decisions according to its own "values", rather than simply executing preset programs. The R&D of "Tongtong" avoids the current data-driven AI's strong dependence on large computing power and big data, and its daily training uses no more than 10 A100 chips, which means that you can learn and grow independently without relying on huge computing resources and massive data. At the same time, "Tongtong" has the ability of intelligent generalization, which can become a universal base for various vertical application scenarios.
Figure: General intelligent people "all pass"
How to measure the intelligence level of "Tongtong"? The world's first general artificial intelligence rating standard and testing platform, "TongTest", provides the answer. TongTest is a comprehensive test system designed to assess the intelligence level of general-purpose agents. Based on developmental psychology and theory of mind, with reference to the test standards for human infant and toddler development, and based on the UV dual system space of "ability and value", TongTest comprehensively evaluates the abilities and values of agents from five ability dimensions such as vision, language, cognition, movement and learning, as well as five value dimensions from personal value to social value to group value. Not only the level of their "intelligence" is considered, but also their "character" is fully examined to ensure that the agent can serve human society safely and responsibly. According to TongTest's staged test of "Tongtong", compared with human children, in complex dynamic scenes, the ability and value of "Tongtong" to deal with complex tasks are comparable to those of human children aged 3-4 years.
Figure: General Artificial Intelligence Rating Standard and Test Platform - "TongTest"
In order to bring the capabilities and value of "Tongtong" from the virtual world to reality, the Institute has developed a general artificial intelligence research platform "TongVerse". As an "AI+robot" simulation training range, TongVerse provides a safe and controllable simulation environment for embodied intelligence, so that robots can be fully prepared and trained before facing real-world challenges. Based on TongVerse, "Tongtong" can migrate to the embodied form in the future, and evolve into tens of thousands of general-purpose agents, out of the laboratory, to thousands of households, and become a trusted partner and helper in people's lives and work. TongVerse has laid a solid foundation for promoting the practical implementation of embodied intelligence technology and the future intelligent industrial ecology.
Figure: General Artificial Intelligence Research Platform "TongVerse"
Establish the heart of the machine and move towards general artificial intelligence
Zhu Songchun, Dean of the Beijing Institute of General Artificial Intelligence, Dean of the School of Intelligence, Dean of the Institute of Artificial Intelligence of Peking University, and Chair Professor of Tsinghua University, delivered a speech entitled "Towards General Artificial Intelligence" at the forum, clarifying what is General Artificial Intelligence and the key path to achieve General Artificial Intelligence.
Dean Zhu Songchun said that the goal of general artificial intelligence research is to seek a unified theoretical framework to explain various intelligent phenomena, and to develop general intelligence with autonomous perception, cognition, decision-making, learning, execution and social collaboration capabilities, and in line with human emotions, ethics and morality. A general-purpose agent needs to have three basic characteristics: completing unlimited tasks, self-defining tasks, and being value-driven. At the same time, it is necessary to solve important issues such as cognitive architecture, value-driven, embodied intelligence, social intelligence, explainability, and human-machine mutual trust.
Zhu Songchun, President of Beijing Institute of General Artificial Intelligence
Dean Zhu Songchun believes that the key to moving towards general artificial intelligence lies in "establishing a heart for the machine." In the past, people mistakenly thought that the ability of human beings to see and do things effortlessly was not intelligence, because these problems were simple, but in fact, it was precisely the strong mind and value system that human beings evolved in response to these problems that were the most important - compared to data and theory, value and "heart" are "alive", and they are the source of autonomous intelligence.
Build an industrial innovation ecosystem and empower thousands of industries to serve China's economic development
In order to accelerate the implementation of general artificial intelligence technology in all walks of life and promote the intelligent upgrading of the industry, this forum officially launched the general artificial intelligence industry innovation ecological plan, which will be based on a series of scientific research achievements and industrial transformation practices of the General Research Institute, with general agent technology as the core, to build an industrial ecology that integrates the four chains of industrial chain, innovation chain, capital chain and talent chain, and promote the development of AI+ new quality productivity.
Dong Le, executive vice president of the General Research Institute, introduced that the general artificial intelligence industry innovation ecology will focus on the six major business areas of emergency, transportation, education, health care, manufacturing, and government affairs, and take the general intelligent human "Tongtong" as the core technology base, and cooperate with ecological enterprises to build vertical field intelligent twins, and accelerate the industrial transformation of general artificial intelligence technology innovation. Based on this, the General Research Institute held signing ceremonies with industrial intelligent twin innovation partners such as Shenzhou Information, China Unicom Smart Footprint, China Southern Power Grid Digital Enterprise, TransInfo Science and Technology Research Institute, and Dark Matter Intelligence, as well as scenario innovation partners such as Weishi Jiajie, Zhizi Engine, Beijing Shudao Intelligent Computing, Tuowei Information, and Aidong Transcendence, to jointly promote the industrial application of general artificial intelligence and help the development of new quality productivity.
Figure: Summit Roundtable Dialogue: Industrial Innovation Practice of General Artificial Intelligence
Focusing on the industrial innovation practice of general artificial intelligence, Executive Vice President Dong Le held a summit dialogue with representatives from all parties in the industry, focusing on the establishment of an open and collaborative industrial ecosystem, discussing the paradigm of cross-field cooperation and innovative business models and application scenarios, providing a platform for the cultivation of original scientific and technological achievements, providing a soil for the growth of innovative enterprises, and a stage for entrepreneurs and developers in the field of general artificial intelligence to realize their potential.
Exploring the boundaries of intelligence, international experts talk about human-machine symbiosis
Focusing on hot topics such as general AI technology innovation, security governance and challenges, Mark Nitzberg, Executive Director of the Center for Human-Compatible Artificial Intelligence (CHAI) at the University of California, Berkeley, shared his views on the future of human-machine symbiosis. Mark Nitzberg said that the ultimate goal of AI development is general AI, but at the same time, the issues of fairness and bias are also important challenges in AI systems, and the symbiotic and collaborative relationship between humans and AI should be explored from an ethical perspective.
Mark Nitzberg, Executive Director of the Center for Human-Compatible Artificial Intelligence (CHAI) at the University of California, Berkeley
剑桥大学生存风险研究中心(CSER)创始主任Seán Ó hÉigeartaigh带来题为"Frontier AI governance on the path to AGI"的报告,Seán Ó hÉigeartaigh认ߐ,人工智能的治理,重要的是在安全和保障等方面达成国际性共识。
图|剑桥大学生存风险研究中心(CSER)创始主任Seán Ó hÉigeartaigh
Building a safe, reliable and efficient AI security governance system has become an important issue for all mankind. Together with Mark Nitzberg, Seán Orange Éigeartaigh and others, representatives of the Institute of General Research and Research conducted a dialogue on the scientific governance of artificial intelligence, and improved the ability of artificial intelligence security governance to ensure the healthy, orderly and safe development of artificial intelligence by promoting the balance between technological innovation and ethics, and establishing an effective risk assessment and response mechanism.
Through "learning, research, production and application", and contribute China's strength to the intelligent era
Towards general artificial intelligence, we need technological breakthroughs in original innovation, and we also need to open up the chain of original innovation, integrated innovation, and open innovation. To this end, the Institute of General Studies, together with Peking University and Tsinghua University, supported the establishment of the Peking University General Artificial Intelligence Experimental Class and the Tsinghua University General Artificial Intelligence Teaching Program. The Institute of General Research will take the lead in carrying out joint training of doctoral students with Peking University, Zhejiang University, Shanghai Jiaotong University and other universities - the talent training plan of the general artificial intelligence collaborative research cooperative. Integrate the respective advantages of the Institute and colleges and universities, and promote scientific research cooperation and exchanges between colleges and universities and the transformation of achievements.
Focusing on the synergy of "learning, research, production and application", the Institute organically combines "free-style exploration" and "organized scientific research" to form a closed loop with the development of China's industry with the new paradigm of "small data and big tasks". In the future, the Institute will work with universities, industry leading enterprises, upstream and downstream industries and other peers to continue to empower China's industrial development with original and innovative scientific and technological achievements, practice "creating the world's science and technology with Chinese ideas", and embark on a general artificial intelligence development path in line with China's national conditions, so as to provide assistance for Beijing's construction of an international science and technology innovation center and a global artificial intelligence innovation source, and contribute Chinese wisdom and Chinese strength to the development of human society.