The Beijing Institute for General Artificial Intelligence (BIGAI; Chinese: 北京通用人工智能研究院; pinyin: Běijīng Tōngyòng Réngōng Zhìnéng Yánjiùyuàn) is a research organization dedicated to advancing artificial general intelligence (AGI). Established in Beijing, China, in 2020, BIGAI receives support from the Beijing Municipal Government and the Ministry of Science and Technology, and collaborates with academic institutions including Peking University and Tsinghua University. The institute is directed by Professor Song-Chun Zhu, who returned to China after 28 years at the University of California, Los Angeles (UCLA).
BIGAI's research philosophy contrasts with mainstream Western approaches that emphasize large language models trained on massive datasets, which it characterizes as "big data, small tasks." Instead, the institute pursues a "small data, big tasks" paradigm, drawing inspiration from cognitive science and developmental psychology. This approach aims to develop what Zhu describes as the "crow paradigm," focusing on reasoning behavior and intelligence based on value and cause-effect relationships, as an alternative to the "parrot paradigm" of current AI systems.
History
BIGAI was founded in 2020 under the leadership of Professor Song-Chun Zhu, a specialist in computer vision, statistics, applied mathematics, and artificial intelligence. The institute was established with government backing as part of China's broader initiatives to advance AI research. Zhu's academic background includes a lengthy tenure at UCLA, where he contributed to computer vision and statistical learning before returning to China to lead BIGAI.
Organization
BIGAI operates as a research organization with affiliations to Peking University, Tsinghua University, and other Chinese academic institutions. The leadership includes Song-Chun Zhu as Director and Founding Director, and Dong Le as Executive Vice-Director. The institute's structure supports interdisciplinary research across multiple AI domains, leveraging collaborations with academic partners to foster innovation in AGI.
Research approach
BIGAI's research approach is grounded in a cognitive science and developmental psychology perspective. The institute argues that current AI systems, particularly large language models, rely on "big data, small tasks" - learning from vast datasets to perform narrow tasks. In contrast, BIGAI aims to develop systems that can learn from "small data" to accomplish "big tasks," mimicking the way humans learn from limited examples and generalize to novel situations. This paradigm emphasizes reasoning, value-driven behavior, and cause-effect understanding.
Key research areas
- Vision and scene understanding
- Cognitive reasoning
- Embodied intelligence
- Multi-agent learning
- Value-driven intelligence
- Autonomous intelligence
Notable projects
Tong Tong AI Child
In January 2024, BIGAI announced "Tong Tong" (also referred to as "Little Girl"), described as an "artificial intelligence child." Tong Tong is a virtual entity designed to simulate behavior and capabilities similar to a three or four-year-old child. According to BIGAI, the AI can assign tasks to itself, learn independently, and demonstrate simulated emotions and value systems, though independent verification has been limited. The system operates on the TongOS2.0 AGI operating system and TongPL2.0 programming language, both developed at BIGAI. The project was presented at the Frontiers of General Artificial Intelligence Technology Exhibition in Beijing on January 28-29, 2024.
In April 2024, BIGAI presented an updated version, Tong Tong 2.0, which it claims has enhanced capabilities more comparable to a 5-6 year old child, including improved language, cognition, movement, learning, emotion, and interaction abilities. These claims have not yet been extensively evaluated by independent researchers outside China.
The Tong Test
BIGAI has proposed the "Tong Test" as an alternative to the Turing test for evaluating AGI. According to BIGAI, the test involves capability assessment across five dimensions - vision, language, cognition, motion, and learning - and incorporates a value system ranging from physiological needs to social values. The scientific community has not yet widely adopted this test as a standard for AGI evaluation.
Philosophy and goals
BIGAI states its mission as "pursuing a unified theory of artificial intelligence to create general intelligent agents for lifting humanity." The institute aims to develop intelligent systems with capabilities in perception, cognition, decision-making, learning, and social collaboration that align with human values. This vision positions BIGAI within the broader global effort to achieve AGI, but with a distinct methodological approach rooted in cognitive science.
See also
- Artificial intelligence
- Machine learning
- Deep learning
- Neural network
- Large language model
- Transformer
- OpenAI
- Anthropic
- Google DeepMind
- Generative AI
- Berkeley AI Research
- MIT CSAIL
- Stanford AI Lab
- Carnegie Mellon University
- University of Toronto
- Oxford University