Domestic scientists "in-depth" is expected to make artificial intelligence "human-like"

author:Xiao Linger talks about technology

A study led by Continental scientists has enabled AI to quickly identify human value intentions, which can improve cooperation between humans and machines in complex work. The findings of this paper have been published in Science... Robotics" in the international journal.

Domestic scientists "in-depth" is expected to make artificial intelligence "human-like"

Research purposes of artificial intelligence

This research is led by Zhu Songchun, a professor at the Beijing Institute of General Artificial Intelligence, which is based on the concept of a framework of human-machine two-way value integration, and demonstrates the instant communication between robots and users to achieve a series of complex human-machine collaborative work.

Let artificial intelligence understand human thinking

Scientists have been trying to let artificial intelligence learn the logic of human thinking, so as to understand human purposes and realize the value needs that people care about. But AI systems, which are now widely used, often do human work in a mechanical way.

What is the "alignment value" of AI?

"If the purpose of a task can be clearly delineated, it is not difficult for AI to understand the value requirements." This is what Zheng Zilong, a co-author of this article, said by Zheng Zilong of the Beijing General Research Institute. However, there are many tasks in life that are difficult to define. For example, a bowl of coffee symbolizes different drinks for different people, and when a person says to an artificial intelligence, "I want a cup of coffee", it must let the artificial intelligence figure out what kind of coffee it wants to drink, that is, it must "align value" with people.

Domestic scientists "in-depth" is expected to make artificial intelligence "human-like"

"Two-way value" artificial intelligence

The research team proposes an explainable intelligent system based on an instantaneous two-way value calibration model. In this system, a group of robots can communicate with people in real time, so as to quickly deduce the user's value goal and "explain" to the user, so that the user understands its value. In addition, the system can guess the user's intrinsic value, resulting in more understandable answers.

Domestic scientists "in-depth" is expected to make artificial intelligence "human-like"


Warm reminder: Beijing Institute of General Artificial Intelligence is a new type of non-profit scientific research institute jointly supported by Peking University, Tsinghua University and other units, and jointly supported by Peking University and Tsinghua University.

Read on