Recently, a joint team led by Zhu Songchun , a professor at the Institute of Artificial Intelligence of Peking University, achieved a major research result, building a computing framework for machines to understand human values in real time, allowing robots and human users to complete a series of tasks through real-time communication. Complex human-machine collaboration tasks. This research can be said to establish a "heart" for machines, allowing AI ( artificial intelligence ) to empower robots to "understand" human values. The relevant paper was published in the latest issue of the international journal " Science·Robotics " under the title "Real-time Bidirectional Human-Machine Value Alignment".
It is understood that the AI system widely used today is a kind of passive intelligence, which can only act mechanically according to the tasks given by humans, and lacks the cognitive and reasoning abilities, emotions and values like humans. In the absence of "heart", it is difficult for AI to understand human beings' true intentions and value needs, and naturally it is difficult to gain human trust and integrate into society. After
discovered the limitations of the previous research paradigm, Zhu Songchun led the research team to change the track, allowing the AI system to independently define tasks in a human way, and "learn" people to start from their own value needs, thereby establishing a "mindset" for the machine. ", allowing AI to learn human value functions, have the ability to align current human value goals in real time, and explore the paradigm of "small data, big tasks".
The research team proposes an explainable AI system based on an instant two-way value alignment model, allowing the machine to infer the value goals of human users through instant interaction with humans and through corresponding feedback, and communicate its decision-making process to the user through "explanation" , to understand the value basis for the robot’s judgment. Research results show that this learning model can improve the efficiency of human-machine collaboration in complex collaboration tasks, thereby enhancing the trust relationship between humans and machines, achieving truly autonomous intelligence, and taking a further step towards general AI.
The copyright of all articles, pictures, audio and video files and other materials reproduced on this website belongs to the copyright owner. If your rights are infringed because the author cannot be contacted, please contact this website and we will take appropriate measures.
