The technology of artificial intelligence is advancing rapidly, prompting the need for continuous improvements in innovative techniques in order to address challenges and seize strategic opportunities in the field, experts said at a recent scientific conference.

“Over the past seven to eight years, AI, particularly exemplified by large language models, has been developed very fast, with ChatGPT achieving 100 million monthly active users just two months after its launch, and DeepSeek reaching 100 million users within two weeks of its launch,” said Dai Qionghai, chairman of the Chinese Association for Artificial Intelligence and an academician of the Chinese Academy of Engineering.

Dai made the remarks in Beijing on Sunday in a keynote speech at the 27th annual meeting of the China Association for Science and Technology.

As a crucial strategic technology and resource, AI is now shifting from horizontal to vertical development, with a focus on advancing more mature technologies to bolster China”s competitive edge in the global technological landscape, said Dai, who is also dean of Tsinghua University’s School of Information Science and Technology.

He dismissed concerns about AI replacing human tasks, noting that current large language models cannot autonomously make decisions, but rely heavily on human input.

Furthermore, due to the complexity and nonlinearity of deep learning models, it is challenging to explain their internal working mechanism, Dai said.

“AI starts from perception of the environment and then uses algorithms to take actions,” he said, adding that sensors are important mediators that transform information from the physical world to the digital one, serving as the foundation of AI. He noted that embodied robots and autonomous driving both rely on vision as well as light detection and ranging technologies, which measure movement and precise distances in real time, for implementation.

At the conference, Dai showed a video of a humanoid robot flexibly performing a somersault and climbing mountains, but failing to put objects on tables when obstacles were in the way. He said that a primary driver of AI, computer vision, is inspired by the feline visual system rather than the human one, so that while it can cope with tasks like positioning and identification, it struggles to address intricate problems and match human-level comprehension.

“The advancement of neuroscience heavily relies on the development of microscopic imaging technologies,” he said, adding that experts at home and abroad are working on this to delve into neural and brain mechanisms, aiming to achieve digital representations that will facilitate a new frontier in AI.

Yu Shaohua, deputy chairman of the China Institute of Communications, said that the increasingly complex and expanding AI system has increased the demand for computing operations, which is straining the capabilities of electrical power capacity. He emphasized the potential of optics to substitute for electricity.

Wang Xiaoyun, an academician of the Chinese Academy of Sciences, stressed that cryptography can help safeguard the privacy of data and information in AI.

In addition, provable security mechanisms can be used to combat deepfake technology — a type of AI that is used to create convincing fake images, videos and audio recordings — and to determine the authenticity of images, videos and other media, she said.