General Artificial Intelligence (AGI) represents a highly demanding field with immense commercial opportunities and ecological needs. Currently, various major model teams are striving for an "entry ticket," resulting in intense industry competition. "From the current layout of AGI, innovation is predominantly on the corporate side, with universities playing a limited role. To transition from following to leading, universities must take proactive steps to leverage their inherent value," stated Dr. E Wei Nan, an academician of the Chinese Academy of Sciences, during a dialogue on the future of artificial intelligence at Shanghai Jiao Tong University yesterday.
To lead the next generation of AGI, universities must address several bottlenecks and focus on strategic points of intervention. Furthermore, they must enhance their capacity to cultivate AI talent and technology effectively. Looking ahead to the next generation of AGI, database technology emerges as pivotal.
According to Dr. E Wei Nan, foundational requirements for developing the next generation of AGI include million-card-scale computational resources, multimodal data resources, AI database capabilities, as well as new model frameworks, algorithms, and engineering capabilities. However, there persists a cognitive misconception regarding the importance of high-quality data, especially in large-scale model research efforts. "Currently, many enterprises and universities are predominantly investing in large model research. However, compared to large models, establishing efficient data processing systems is a more pressing task," Dr. E Wei Nan emphasized.
He pointed out that while the importance of high-performance chips is widely recognized, there's an unexpected challenge: "From a technical perspective, chip performance metrics are easily quantifiable, whereas data quality is not." Across various sectors such as research, healthcare, and finance, effective data analysis and processing are indispensable. Therefore, database technology stands out as one of the critical factors in developing the next generation of AGI.
Specifically, robust AI databases are essential for data collection, cleaning, evaluation, extraction, and other preprocessing tasks. "Only by establishing robust databases can we promptly address any deficiencies during large model training and maintenance," Dr. E Wei Nan elaborated. Currently, both domestic and international enterprises have started focusing on AI databases and initiated a new round of competition. For instance, the first international AI database, MyScale, developed by Mochi Technology, whose founder and CEO, Tai Cheng, and CTO, Tang Linpeng, are graduates of Shanghai Jiao Tong University, is currently the most comprehensive and powerful AI database in Dr. E Wei Nan's view.
Intellect combined with computational power: universities and enterprises breaking barriers together. At the dialogue, experts emphasized that two major elements are crucial for developing general artificial intelligence: intellectual resources, primarily concentrated in universities, and computational resources, mainly held by enterprises. However, the current challenge facing AGI development in China lies in the lack of effective integration between these two resources. "While some teams have gained a competitive edge by leveraging their engineering capabilities and computational resources, this is unsustainable in the long run as industry standards continue to rise," Dr. E Wei Nan stressed.
He emphasized that for universities to secure a position in the next AI frontier, they must strengthen organized research efforts and collaborate with enterprises for effective resource integration and alignment.
According to a study published in Nature, the pace of breakthrough scientific innovation is slowing down. Professor Zhou Bowen from Tsinghua University's Hui Yan Chair added that the formation of information silos, information overload within these silos, and excessively high barriers between them are underlying reasons. Therefore, both the organization of scientific research and the proactive engagement of researchers must evolve.
In response to these challenges, Ding Kuiling, President of Shanghai Jiao Tong University and an academician of the Chinese Academy of Sciences, expressed deep concern. According to the university's research, over 20% of its faculty and researchers, nearly 750 individuals, are currently engaged in AI-related research.
Regarding the newly established Shanghai Jiao Tong University AI Institute, Ding Kuiling outlined its future strategy. The institute aims to aggregate talent, foster organized research, and enhance parameter volume. It also intends to tackle real-world problems, serve real scenarios, and actively respond to the demands of academic disciplines, industries, and socio-economic development to enhance data volume. Furthermore, by consolidating resources and ensuring computational support, it aims to enhance computational volume.
