New Work of Zhu Songchun Team Dominates Headlines

Paper link：https://www.science.org/doi/10.1126/scirobotics.abm4183

This paper proposes an interpretable artificial intelligence (XAI) system, elaborates on a computational framework for real-time understanding of human values by machines, and demonstrates how robots will communicate with human users in real-time to complete a series of complex human-machine collaborative tasks. Zhu Songchun Team has been engaged in interpretable AI related work for a long time. The article is the second paper of the team on interpretable artificial intelligence published in Science Robotics. The research covers cognitive reasoning, natural language processing, machine learning, robotics and other multi-disciplinary fields, which is a concentrated reflection of the cross-research achievements of Professor Zhu Songchun team.

To Establish "Perception" for Machines

An important Step to Implement "Small data, Big Task" Paradigm

The artificial intelligence system widely applied today is a passive intelligence. It can only mechanically follow human given tasks, lacking cognitive and reasoning abilities like humans, as well as emotions and values. In the absence of "Perception", artificial intelligence is difficult to understand and execute the real intentions and value needs of human, naturally making it difficult to gain human trust and integrate into human society.

After discovering the limitations of the "Big data, Small task" Paradigm, Professor Zhu Songchun team switched tracks to explore the paradigm of 'Small data, Big task'. The core of 'big task' is to allow artificial intelligence system to autonomously define tasks according to human ways. As tasks defined by human starts from their own value needs, it becomes the core work of "small data, big task" to establish "perception" for machines and enable artificial intelligence to learn human value functions.

The release of this research achievement marks the ability of artificial intelligence systems to learn human value functions in real-time communication and align current human value goals in real time. It is an important step to achieve the paradigm of "small data, big task".

Upgraded Human-machine Collaboration

Driven by Value Consensus to Activate AI Potentials

From the perspective of industrial practice, it is no longer a novelty for the industrial application of artificial intelligence technology. However, AI still remains at a relatively junior and low value link in the entire industrial value chain, i.e. to implement repetitive basic work trained on a large amount of data based on fixed rules and simple tasks set by humans. It is truly a "tool man". The constantly boasting massive computing power, algorithms, and data in the industry intensify the "rat race" of the low value dimension.

If to achieve value upgrading in the industrial application field, the artificial intelligence technology must understand the global value intention and achieve value alignment through dynamic interaction with collaborative members to drive the execution and implementation ultimately. The research achievement of Zhu Songchun team has made the industry foresee the widespread applications of AI from "low dimensional data driven" to "high dimensional value driven".

As an important practical carrier of strong cognitive artificial intelligence technology, DMAI inherits the disruptive "small data, big task" innovation research paradigm. By building a strong cognitive AI technology platform, it has been implemented and applied in multiple fields of intelligent social governance, industry transformation and upgrading.

In the field of urban governance, DMAI empowers cognitive AI capabilities, such as trend analysis, cognitive reasoning, value deduction, and decision-making assistance in urban management, achieves the value upgrading of artificial intelligence from junior perception to cognitive reasoning and decision-making assistance, and further improves AI efficiency and reduces governance costs. In the field of smart commerce, DMAI launches a digital virtual human, enabling the underlying AI technology of the virtual human stack through machine vision, speech recognition, natural language processing, and cognitive dialogue system. In terms of retail scenes, DMAI builds rigid demand service applications, such as virtual anchors and virtual customer service, to achieve human-computer cooperation based on value consensus.

With strong cognitive AI as the core, DMAI will continue to build a new innovation chain that integrates science, research, production, and application, and accelerate the transformation of cutting-edge scientific research achievements into new driving forces for industry transformation and development. Thus, DMAI will further activate the industrial value of artificial intelligence, serve national strategies and improve human well-being.