Huang Yufeng, a nationally renowned Chinese language education expert and a special Chinese teacher at Fudan Affiliated Middle School, has been committed to educating students for many years, but recently he has encountered a new problem - in the era of artificial intelligence, humanoid robots will increasingly enter the family, the classroom, and the society. Do we still have to be a humble gentleman in the face of "it"? Should we also let "it" abide by the minimum gentleman's way? Such questions aroused the curiosity of many participants at the 9th Gentleman Cultivation Conference held a few days ago.

AI doesn't understand the self and the self

A few days ago, Huang Yufeng, who is currently the principal of Fudan Wupuhui Experimental School, asked two small questions to the big language model when preparing for Su Shi's "The Fu of the Later Red Cliffs" with the last sentence in the article "The Taoist priest smiled, and I was also shocked. Open an account and looked at it, but I couldn't see it." He asked two small questions to the big language model: one is what "shock" means, and the other is who "shock" means. "I didn't expect that the answers of artificial intelligence are inaccurate. For example, the explanation of the word "一" obviously cannot recognize the original meaning of the article. It neither understands the "一" represents the "me", nor can it understand and express the philosophical propositions such as "where I come from and where I want to go." Huang Yufeng said that an important feature of a gentleman is that he is good at learning, especially the spiritual literacy of "learning and determined, asking questions and thinking closely", but whether it is AI search tools or humanoid robots, they do not have the qualities of gentleman in "一", "一" and "年". A robot that does not understand the ego and the big self and does not have confidence is ultimately just a tool used by humans.

"Everyone's Teacher" is just a "teacher assistant"

"Humans are elevated to the moral level from various behavioral practices, while robots rely on instructions input by humans to act and think. They need to rely on a large number of examples to learn human moral norms. This is the difference between humans and machines." Wu Gang, a professor at the School of Education of East China Normal University, said that humanoid robots will not have the unique warmth of humans. Therefore, even if they can walk into the classroom in large quantities in the future, they will only be a "teacher assistant" at best, similar to a "all-knowing teacher" with a large database. It will tell students the ready-made answers and output existing knowledge, but it cannot replace the main teacher's responsibility to teach. Because the ideological confrontation, cordial heart-to-heart and emotional interaction between teachers and students and classmates cannot be achieved by any large language model database.

Ethics need to be "two-way"

Yang Xiong, director of the Shanghai Children's Development Research Center, raised a warning: Don't randomly input large language models such as bean bags and other valuable problems. If all you feed them is messy, AI will also have "illusions" after continuous reception, thinking that "this should be the case." For example, when many people ask it whether to buy a certain stock, it is likely to think that the stock that everyone is asking about should be "high-performance stocks". Therefore, when people interact with AI, a basic ethical norm must be followed, which is to maintain positive and advanced dialogue content and prevent it from being "serious and nonsense". Humanoid robots are just plug-in learning tools and cannot become a critical gentleman.

Some experts at the meeting said that developers should incorporate transparency and verifiability mechanisms in every link of model training and algorithm deployment to prevent the generation and spread of false information from the root. For example, in the data collection stage, it is necessary to clarify the source and purpose of the data to ensure the authenticity and reliability of the data; during the model training process, interpretability algorithms must be used to enable the model's decision-making process to be understood and supervised. Once false information is generated, it can be traced back to the link where the problem arises. At the same time, relevant laws and regulations should be accelerated to punish acts of using AI to create and disseminate false information. This requires humans and artificial intelligence tools to complete a "two-way rush" in basic ethics.

"When humanoid robots begin to enter our real life and real work scenarios, they need to reset the ethical norms of the entire society in advance, so that AI tools must also abide by human public order and good customs." Lu Mingjie, associate researcher at the Intelligent Social Governance Laboratory of Zhejiang Zhijiang Laboratory, said that taking smart cars as an example, once an accident occurs during driving, how should we determine the responsibility? Therefore, intelligent tools such as humanoid robots are facing ethical reset and reshaping, which they cannot avoid and are also problems that we urgently need to pay attention to and solve.