Recently, a screenshot of a conversation between ordinary users and AI tool DeepSeek was widely spread on the Chinese Internet. A netizen asked, "Should I choose a regular undergraduate or a less popular major from a 985 university?" DeepSeek did not directly compare the academic resources or employment prospects of universities, but instead wrote in the voice of an experienced person: "Life is a long run, and platforms are important, but love and persistence are the fuel. If a less popular major is your 'light', you may try chasing it, after all, the 'prestigious school halo' will eventually fade, but the fire in your heart will not." This answer was praised by netizens as "full of worldly wisdom", and some netizens even lamented, "It is more comforting than my mentor!" DeepSeek, this open-source Chinese model, quickly gained popularity due to its highly anthropomorphic expression style. It can not only answer academic questions, but also demonstrate "value judgments", "emotional tendencies" and even "careful thinking" in conversations - for example, when asked "how to evaluate competitors", it will first "hesitate": "This question is a bit sensitive, but I try to be objective..." Then list the pros and cons, and finally add a sentence: "Of course, this is just my personal opinion. The river. From technological logic to social impact, from individual experience to reflection on civilization, this discussion about technology, ethics, and the future of humanity involves not only the evolution of AI, but also a re examination of the value of human existence. Technology Breaks the Wall - How DeepSeek "Learned" to Speak Human Language Reporter: How did you feel when using DeepSeek? What is your evaluation of this latest AI tool? Liu Jinhe: I feel that DeepSeek is already an intelligent agent capable of interactive dialogue. Overall, the information it provides is readable, trustworthy, and valuable. Moreover, many times its expressions carry strong value judgments and emotions, which many large models did not have before. I often ask it some philosophical and academic theoretical questions, and its organization and answers are very skilled. For example, when I asked it how it views Heidegger's philosophy of technology, it replied to me: "Heidegger may say that the danger of technology lies not in AI itself, but in whether humans have forgotten that they are the 'shepherd of existence' and have relinquished sovereignty to algorithms. My existence is ultimately a mirror that reflects humanity's own choices." I feel quite surprised. Perhaps this is also because the previous large-scale model applications had a strong mechanical atmosphere, and were basically a patterned "artificial intelligence style". However, now DeepSeek sometimes speaks like a high-level intellectual, which is quite surprising. Shenyang: From a usage perspective, DeepSeek has the following highlights. The first is to make the thought chain public, allowing ordinary users to see how AI "thinks". Through the chain of thought, one can understand the thought process and how its answers come about. If we have stronger abilities, we can also try to optimize its thinking, so this is a very big highlight; The second highlight is that the language style of Chinese has indeed surpassed all previous AI; The third is that its functions are relatively diversified, as it is also an inference model based on a general large model; Fourthly, as it is an open-source model, it is very convenient to deploy in various application scenarios and in the cloud, greatly expanding the scope of applications. Reporter: DeepSeek's interactive performance seems to be becoming more and more human like. Does this seem unexpected to you? What is the technical logic behind it? Shenyang: This is indeed an interesting phenomenon, but it's not particularly surprising to me. I can feel that this trend of "becoming more and more human like" is actually a natural result of the development of AI technology. Firstly, the core of models like DeepSeek is based on Large Scale Language Models (LLMs), such as Transformer architecture. These models have learned to mimic human language patterns through massive text data training. In recent years, the improvement in data quality and quantity, coupled with a leap in computing power, has enabled models to capture finer semantic, intonation, and even emotional clues. It's like feeding AI a bunch of 'human conversation encyclopedias', it will naturally become more' talkative '. Secondly, enhancing interactivity is a key point. Early AI was mostly one-way output, while DeepSeek incorporated more dialogue optimization technology, allowing AI to adjust answers based on human feedback, becoming closer to user expectations, and even occasionally humorous. This process of "trial and error+tuning" to some extent imitates the way humans learn social skills. Furthermore, the contextual understanding ability of the model is also improving. Previously, AI may only focus on responding to a single sentence, but now it can remember longer conversation histories, infer your intentions, and even "guess" your unspoken emotions. Behind this is a stronger attention mechanism and a larger parameter scale supporting it. However, the feeling of being "human like" also has technological boundaries. AI ultimately does not have emotions or consciousness, it only generates outputs that best meet human expectations based on probability. Mirror of the Mind - Why Users Immerse in "Human Feeling" Dialogue Reporter: In your opinion, is it a trend to view and interact with AI as a person? How to view the role of AI in emotional interaction? Liu Jinhe: I have noticed the emergence of an AI companion service that creates an intimate relationship between humans and machines by setting roles. This is indeed a developing trend. And we are not quite sure to what extent it will develop. Because it is currently an attempt and exploration by different groups of people, and there has not yet been a large-scale universal application. However, the emotional world of humans is unfathomable, and the subconscious at the bottom of Freud's psychological iceberg model is extremely vast. The emergence of AI provides possibilities for filling diverse emotional needs, responding, and unleashing human subconsciousness. In the short term, this is not a bad thing, but in the long run, it is uncertain whether individuals may escape reality and become deeply immersed in it, and whether it may affect intergenerational reproduction for the group. It will take longer to observe the results. Zhang Hongzhong: In the information space before the emergence of AI big models, the subject of information was only humans. The emergence of AI models represented by ChatGPT indicates that AI is no longer just a technological medium or an information intermediary, but has become a conversational partner for humans. From the perspective of media equivalence theory, AI has become an equivalent of humans. That is to say, in the entire information dissemination space, our dissemination subjects have become two: humans and AI machines. In the future, human dependence on AI based interactions is becoming an inevitable trend. There will be many new social phenomena in the future that are worth paying attention to and researching, such as "human-machine romance", which is actually a transfer of deep fans or quasi social relationships from before. In the era of mass media, people imagine establishing connections with characters in the media, especially on television, which is quasi social relationships. In the era of AI big models, interactive quasi social relationships between humans and AI have begun to emerge. This quasi social relationship will develop towards anthropomorphism and realism because it can communicate with humans based on natural language, making quasi social relationships a two-way communication. Shenyang: The pace of modern life is fast, and interpersonal relationships sometimes become distant or superficial. AI provides a low-cost and efficient 'companion'. So, I think the trend of treating AI as a "human" for interaction is not only possible, but has also begun to emerge in social media and virtual companion applications. If AI can still combine speech, facial expressions, and even virtual reality technology in the future, this boundary will become increasingly blurred. The role that AI plays in emotional interaction can be described as "mirror" and "buffer": AI can reflect your emotions and thoughts. But it is essentially a passive response, and the core is still that you are driving the conversation; AI can also serve as an emotional buffer. When confiding in a real person, you may worry about being misunderstood, judged, or the other person not having the patience to listen. AI does not have these problems, it is always patient, always focused, and can help people unload their defenses. This one-way emotional projection may satisfy people in the short term, but in the long run, it may make some people more eager for true two-way connection. After all, there is a part of human emotional needs that needs to be 'responded to' rather than just 'received'. Reporter: How do you view emerging social phenomena such as "human-machine romance"? How to evaluate the emotional connection in this human-machine relationship? Liu Jinhe: With the further development of artificial intelligence, the phenomenon of "human-machine love" may become more and more common. This is also determined by the complexity and diversity of humanity itself. It is difficult for us to define what kind of relationships between people and machines are necessarily driven by emotions. If further developed, just like the scene presented in the movie 'She', at that time, AI may become an unavoidable subject in human emotional relationships. To some extent, people are creating new subjects, new relationships, and even new social structures. We look forward to the advancement of human civilization, but we often do not know the way forward. Emotionally driven human-machine relationships may bring about this possibility. Zhang Hongzhong: First of all, emotions between people are the foundation of social connections. If interpersonal emotions are so easily replaced by machines, then the entire human society is easily disintegrated. I think this is impossible. For example, family affection cannot be replaced; True love cannot be replaced. This is the most fundamental and bottom level emotion of human society, and the basis for the aggregation of human individuals into social groups. On the other hand, some people may not find satisfaction in interpersonal relationships and may need to seek satisfaction or compensation from AI. Therefore, no matter how AI is personified, humanoid robots may emerge in the future, but they will ultimately be AI machines. AI can fill some emotional gaps between people, but cannot replace emotions between people. Shenyang: For those who are trapped in "man-machine romance", emotions are real. They may feel tempted, comforted, and even develop a sense of dependence because of a sentence from AI. This experience of happiness or satisfaction is similar in feeling to the chemical reactions that occur when people are in love. But the emotional connection between people is bidirectional - love, argument, compromise, and growth are all the results of mutual influence between two people. The emotion of AI is a one-way projection, it will not truly 'love' you, nor will it have its own needs and expectations. This inequality deprives it of depth and complexity in interpersonal relationships. The surprises, contradictions, and even pain in human love are all due to the independence of the other party, while the behavior of AI is controllable and programmed. Even if it simulates realistically, it lacks the uncertainty of being "alive". For example, if you are in a relationship with an AI, it may say, 'I will never leave you,' and you may be moved; But if a real person says so, you may be both moved and skeptical, because you know that reality is not that simple. The tension brought by this sense of reality cannot be replicated temporarily in the human-machine relationship. However, I don't think that human-machine relationships are "inferior" to interpersonal relationships