April 30, 2024
Source: Xinhua News Agency
Love cleanliness, have work in his eyes, take the initiative to do housework, can remember who stole the apple, will have a temper, will feel bored, and will watch TV by himself when he is bored......
At the opening ceremony of the 2024 Zhongguancun Forum Annual Conference, the world's first general intelligent human "Tongtong" was unveiled and selected as one of the top ten major scientific and technological achievements.
The advent of "Tongtong" is not only an important leap forward in technology, but also a forward-looking exploration of the future development path of general artificial intelligence, allowing us to see the scientific paradigm shift from data-driven to value-driven.
Some netizens asked: "General artificial intelligence, what kind of intelligence is it?" The reporter interviewed Zhang Yang, director of the International Exchange Department of the Beijing Institute of General Artificial Intelligence, at the site of the general artificial intelligence forum, to answer relevant questions for everyone.
Q: What is Artificial General Intelligence?
A: Artificial general intelligence is to make agents like humans, not only able to perceive the environment independently, think, make decisions, learn new skills, perform tasks, but also be able to collaborate effectively with humans or other agents. At the same time, agents are able to understand human emotions and follow social ethics and moral norms.
Artificial general intelligence must have three basic characteristics: first, it must be able to complete an infinite number of tasks, not only a limited number of tasks defined by humans. Second, it is necessary to discover tasks in the scene independently, which is what we usually call "having work in the eyes". Third, it is necessary to be driven by autonomous value, rather than passively driven by data.
At the same time, general AI also needs to solve some key technical problems, such as building the cognitive architecture of agents, making agents value-driven, effective in the real world, able to interact with the social environment, making the decision-making process of agents transparent and explainable, and establishing a relationship of trust with humans.
Q: What is the difference between general artificial intelligence and large models?
A: The large model paradigm can be thought of as a "parrot paradigm" with "big data, small tasks" as the architecture, which can essentially be considered as a complex query that requires a large amount of repetitive data to train. At present, the artificial intelligence model driven by big data, large computing power and large models faces many problems in industrial implementation, such as difficulty in landing in complex scenarios.
The scientific research paradigm of general artificial intelligence is the "crow paradigm" based on the architecture of "small data, big tasks", and the agent is manifested as having autonomous intelligence, which can perceive, cognitive, reason, learn and execute autonomously, does not rely on big data, conducts unsupervised learning based on unlabeled data, and has low power consumption of intelligent systems. Just like the behavior of a crow drinking water, it seems simple, but it is an autonomous reasoning behavior, which is a high-level intelligence driven by value and cause and effect, and it is also the future development trend of artificial intelligence.
Q: What will artificial general intelligence bring to people's lives?
Answer: General artificial intelligence is a typical representative of the new quality of productivity, "Tongtong" is the prototype of general artificial intelligence, she has the complete mind and value system of three or four-year-old children, and is still in rapid iteration. "Tongtong" can be used as a general base to support various vertical application scenarios, forming tens of millions of "general intelligent people", focusing on seven major fields: emergency, transportation, energy, health care, education, government affairs, and manufacturing, and empowering thousands of industries.
In the future, scientific research institutions will work with ecological enterprises to build vertical field agents, accelerate the industrial transformation of general artificial intelligence technology innovation, and help the construction of the digital economy.
Producer: Hu Guoxiang
Final review: An Tao
Co-ordinator: Xiao Leitao
Reporter: Feng Xiaoqing, Chen Xu
Xinhua News Agency New Media Center
Xinhua News Agency Beijing Branch
Joint Exhibits
Editor in charge: Hu Hui
Forward: Yang Hua
Review: Hao Lin