Beijing Institute for General Artificial Intelligence (BIGAI)

What it is

BIGAI, the Beijing Institute for General Artificial Intelligence, describes itself as a non-profit "new-type" research institution. Its own account says it operates under the guidance of the Beijing municipal party committee and government and was promoted by the Beijing Municipal Science and Technology Commission.^[2] The institute says it works with Peking University, Tsinghua University and other institutions under the joint support of Beijing, the Ministry of Science and Technology and the Ministry of Education.^[2] Its address is the Future Science and Technology Building, 2 Yiheyuan Road, Haidian District, Beijing.^[2]

The name points at the ambition. Most industrial AI systems are built to do one narrow job well — recognise faces, translate text, recommend videos. General artificial intelligence, as the institute frames it, is the attempt to build systems that hold a broad range of cognitive abilities together rather than one skill at a time.

Who runs it and where it came from

Zhu Songchun is named as president. He holds at the same time the titles of chair professor at Peking University and chair professor of basic science at Tsinghua University.^[2] In his message on the institute's site, Zhu writes that he returned to China full time in September 2020 after 28 years of studying, living and working in the United States, and set up BIGAI.^[2] A Peking University profile of Zhu describes BIGAI as a sister institute located near Peking and Tsinghua universities.^[1]

That geography matters for how the institute is organised. Rather than standing apart from the universities, it is set up to sit alongside them, drawing on their faculty and students while running its own research programmes.

What it works on

The institute's stated task spans six key fields: computer vision, natural language, machine learning, cognitive reasoning, robotics and multi-agent systems.^[2] These are the components that a general agent would need in order to perceive, understand language, learn, reason about the world, act physically, and cooperate or compete with other agents.

BIGAI states a goal of pursuing a unified theory of artificial intelligence in order to create general intelligent agents.^[2] The word "unified" is doing real work here: the aim is not a collection of separate task-specific models but a single theoretical account that ties perception, language, learning and action together. That is a considerably more demanding target than improving any one benchmark.

To support this, the institute's task list includes building an open "big task" platform. The platform is meant to provide physically and socially realistic large-scale training and testing systems, at both the individual scale and the social scale.^[2] The emphasis on social scale reflects a view that intelligence in the everyday sense is not a solo activity — much of it shows up in how agents coordinate, negotiate and share intentions.

Teaching

BIGAI pledges to work with Peking University, Tsinghua and other universities on a full artificial intelligence curriculum spanning undergraduate to doctoral level. The stated aim of that curriculum is to produce generalist talent oriented toward general intelligence.^[2] This is a deliberate contrast with curricula organised around individual techniques and tools: the institute is arguing for training students who can think across the boundaries between vision, language, robotics and reasoning.

Place in the field

BIGAI belongs to a group of organisations founded on the premise that the next stage of AI is not more narrow applications but machines with broad, integrated capabilities. Its distinguishing features are its non-profit, municipally backed structure, its formal ties to two of China's leading universities, and its declared intention to work from a unified theory rather than from a portfolio of separate systems.^[2] Whether such a theory can be built is an open research question; the institute's six-field scope and its platform-building plans describe how it intends to try.