Artificial intelligence not only becomes an important driving force leading to a new round of scientific and technological revolution and industrial revolution，and profoundly affects knowledge production and communication including social sciences。New development of social science research around the era of artificial intelligence、New Trends and New Challenges，"News of China Social Sciences" invited about Micharine、Gao Qiqi and Qiu Lin three scholars start a discussion，I hope to promote consensus through dialogue、Show differences，Spark of Social Science Research in the era of change。

　　Bet365 app download

"China Social Sciences": We said that the appearance of ChatGPT in the history of artificial intelligence has a milestone。If a word is used to describe the essence of ChatGPT，or revolutionary breakthrough，What? We can see the useful generation artificial intelligence、GM artificial intelligence、Big language models and other different expressions，What is the difference behind it? 

Qiu Lin: I advocate using a big model to understand ChatGPT。Why not "generate"? Because of this language model，Actually, from the beginning of AI。The earliest Eliza，It is a dialogue type，But it is based on very simple rules。For example，You said "I'm in a bad mood today"，It will dig out the keyword "in a bad mood"，Generate "Why are you in a good mood?" Although this conversation is very stereotyped、Mechanized，But it is also essentially generated，It is based on your input，Reaction to you。Artificial intelligence is a very surprising tool when the dialogue robot just developed a dialogue robot，Because it was originally used for psychotherapy，Chatting with dialogue robots in psychotherapy，Let people express more。So say，Generation formula has a certain history for artificial intelligence，Not new。

The reason why it is called a big model，Because ChatGPT's current data is the largest in history，Add its amazing computing power，This explosive intelligent behavior。The biggest difference between it and previously generated artificial intelligence is，Its training set is oversized，And the algorithm behind it is also different from before，is a machine learning based on neural network，This is a more intelligent self -learning。

Miganin: The generated artificial intelligence is very early，But the generation formula of ChatGPT is very unique。Why do you have no need to generate artificial intelligence before? That's because the quality of the previously generated content was poor，Although the content was generated，but not smart enough。until today，Genetic artificial intelligence can really reach the degree that can be used。

The full name of GPT is the Generation Pre-Trained Transformer，Translation is a generative pre -training converter。Add CHAT，I think Open AI is out of business considerations。But put it in Chinese，It is easy to understand it as a chat robot，Even the use effect is not as good as ordinary chat robots。Actually，GPT is a collection of three elements: First, big model，The second is large computing force，Three is the big corpus，Three missing one。In this whole，Macrometer is the core，How much computing power is because there is，How much corpus can be processed; how much computing power is，How much model can be established。In the Sino -US Science and Technology War，Why does the United States suppress us from the chip? Because the chip is the basis of the computing power，Control the chip，The computing power will not come up。

Gao Qiqi: I agree to define ChatGPT from the big model。From the perspective of conceptual history，Genetic artificial intelligence This concept is slightly earlier than the big model。Three or four years ago，The concept of more discussion is AIGC，that is, AI Generated Content，The content generated Bet365 lotto review by artificial intelligence automatically created。This concept has two related concepts，One is PGC，Professional Generated Content，It is a professional content; one is UGC，User Generated Content，It is a user generating content，That is the content of generating。AIGC concept is earlier than big models，Chatgpt after appearing，In order to better define，People are increasingly inclined to use big models。Actually，More accurate saying is the big language model。This language is 4 basic categories about language from artificial intelligence -machine vision、voice generation、Natural language classification of natural language and robots grows。

Slightly supplemented，This big model is not only a large amount of data，Parameters are a more essential feature。Because the most important feature of this algorithm is to imitate neural network。GPT-3 parameters are 175 billion，GPT-4 officials did not announce，But what you can know is，It is a hybrid expert model (MOE，MIXTURE of Expert)，It is a model combined by different small models。So，Early artificial intelligence is based on rules，is also symbolism; and developing to a neural network is actually associationist，The current big model is more reflected in correlationism。

　　Not only new tools, but also a new paradigm 

"China Social Sciences": in recent years，With the development of new technologies，Academic circles, especially the public management community, a voices related to the transformation of the research paradigm，For example, a new research paradigm that advocates building data drivers。The emergence of a big model，What new possibilities have you opened up to social science research? 

Migin: Before the appearance of the big model，The research paradigm of social science mainly has qualitative research、Quantitative research、Simulation research and big data research。The proposal of big data is in 2010，It has a great impact on social sciences after it is proposed。Before that，Social science research in the past is based on individual research on problems。Kong De has always advocated that social sciences are aligned with natural sciences，He much respects the study of classic physics in nature，I hope that social science can also find some rules like physics。Followed by Kong De，Social sciences began to conduct a large number of quantitative research imitating natural sciences。but，These studies have not made any progress。Because，Research objects of social sciences are super complex，Research by a regular method，equivalent to using simple quantitative methods to cope with complex problems，It seems to use the method of decomposing a factor to solve the problem of the ball points。Actually，Quantitative research is only possible in very small fields，For example, company governance，Because the company's border is clear、Data specification，You can make some regression analysis。

The complexity of social sciences is more about emerging problems。For example，Inflation in the United States，Why does the Federal Reserve do not need to measure economics to make a model to solve it? Because it knows，The emergence of inflation rate is the result of the interaction between the entire American society in the real world，Including consumers、Bank、Interaction between producers mutual。

After the big data appears，Everyone thinks that big data can have some solutions，Want to observe people's behavior through big data，but not successful。Big data has never formed a systematic system。Today，Chatgpt is here，First is the change of epistemology，Then the change of methodology，The solution to the ground is here。Because it was a problem -based research in the past，It is an event -based research; and ChatGPT is based on the overall wisdom of human beings。The transformation of epistemology is your social system、The understanding of complexity has changed。For example，Qualitative Studies、Quantitative studies are super simplified the complexity of society，Simulation research seems to want to simulate the complexity of society，bet365 Play online games Even combined with big data to engage in digital twins，But they can't really solve the problem。ChatGPT brings a generative -based governance solution，Timely、Flexible、Quick。Why does it say it is generated? for example，In the field of emergency management，In the past, we did emergency management，The first is to find out the emergency plan，Take step by step according to the emergency plan。but，Facing super complexity、Uncertainty，How could there be a plan? Actually，During the emergency management process, it is often solved according to experience，But experience is limited after all。ChatGPT is here，You can give a decision support system for every scene、Decision Support Plan，It can allow decision makers to refer to the solution given on the basis of experience，and the progress of the crisis，Constantly generate new solutions。

Big model as the fifth research paradigm，Definitely going out of a new road。Before big data，Social science solution is driven by theoretical，Big Data as the fourth research paradigm，Data -driven social science research was formed。With a big model，The most needed is to liberate the thinking framework。Previously a problem -based solution，Later will be an event -based solution，Not from the perspective of economics，Not from the perspective of political science，How to solve it when it is easy。

Gao Qiqi: Quantitative simulation、Big Data、The big model is actually a different stage of development，Methods used by computer science。In the final analysis, they are all quantified methods，is quantitative。The development of big models will definitely make this quantitative method easier，Because it is trying to capture all the possibilities。But this is theoretically impossible，Because everything will produce new variables。If you really want to exhaust，Then enter a closed system。

Qiu Lin: Many of the original social science research samples are relatively small，Like a psychological experiment, 100 people may be divided into two groups，The control group and the control group ratio，Discover differences，Data in actual life，For example, shopping、Eat、Transportation is very difficult。Big data brings a disruptive change，Because it allows us to get a large amount of data，For example, you can measure emotional changes within 24 hours through social media。These real -time、Large amount of data，Make up the problem that the sample volume is too small in the original social science research。Sample amount is too small，It means that the generalization ability of research is small，Big data has expanded our vision and data amount，Let us study the phenomenon that we can't study before。This is the advantage brought by big data。

Now talk about big models，different from big data。For me，The big model is a subversive tool。It can make social scientific research producing a qualitative leap in the breadth and depth。For example，According to recent reports，Researchers let ChatGPT produce a robot that can be walked。The robot imagined by human beings is often similar to human or animals，This limits our thinking; and the answer given by ChatGPT is beyond human imagination -it designed a sponge，There are many holes above，Move by cheering。It can be seen that it has strong creativity。This will greatly promote theoretical innovation。Actually，ChatGPT can be scheduled from the literature、generate theory、Research Design、Collect data、Analysis of data and other aspects to empower social science research。

Big model's understanding of human society，Not a separate understanding，but an understanding of the global view，Is a systematic understanding。Use big data to do research，For example, analyze people’s emotional changes based on data on Weibo，or look at the situation of traffic congestion、The law of travel，These studies are still based on some data to describe human behavior，The theory produced in this way is still more isolated。and real life is a very complicated system。After a big model,，It can generate an analog bet365 live casino games system，to simulate very real individuals and group behaviors。Smart town established by the research team of Stanford University，The 25 intelligences in it can produce very credible、In this town。The individuals in the past computer simulation are very simplified，Set behavior in a very mechanized way。In this town，You can tell one of them you want to organize a Valentine's Day party，This person will tell others about the time for partying，Other people in the town will make up for each other，Who will go with who with whom，Remember time，and appear at the accurate time and place。They are no longer limited to the simple behavioral rules of the established in the modeling，Instead, a large model generates corresponding behavior behind the interaction。This interaction scene similar to real life in a virtual environment，It provides great convenience for the research of individual and group behaviors for researchers in social sciences。

Our recent research also shows，The ability of a large model to simulate the ability to generate large -scale survey data。In March 2023, the US Petou Research Center launched a questionnaire survey on the view of the Russian and Ukraine's conflict between Russia and Ukraine and the Bynden government's response.，10441 interviewee，The survey contains 8 questions。We started a study in June 2023，Research Design is to introduce to ChatGPT to the questionnaire question of the Russian and Ukraine Research Center in March 2023，Then it is required to predict the percentage of American interviewees on each issue of each issue。Research indicates，In 8 questions，Chatgpt's prediction on 7 issues is consistent with the direction of the selection of the survey results。We also did the same experiments on the polls of another 3 other Pippi Center，A similar results have been obtained。This description， The big model has mastered people's psychological state and way of thinking，It will be tested by humans in social science research。

"China Social Sciences": If a large model such as ChatGPT has the ability to simulate interviewees to generate large -scale data surveys，Is there any room for research methods such as questionnaires? 

Migin: I want to emphasize the change of epistemology，The generation of knowledge is from the generation of traditional individuals to the generation of human knowledge based on human knowledge。The most critical place here is，is the power of the individual，I still believe in the power of the big model。Our past social science research is based on individual power。I think the power of the individual is insignificant。In front of the big model，The individual exists just as a catalyst。Future research lies in，Who is better at talking to big models，Then the solution it generates is better。Consider assumptions、Data、The model is not necessary，Chatgpt can be solved。Just handle the context，The questionnaire of the individual can be fed as a small corpus to a large model。The content generated by the big model must be more reliable。So for polls，Not allowing ChatGPT to go with the polls，Instead, it should be used to see the polls and are not allowed。

Qiu Lin: The results of the simulation of the big model are really consistent with the results of the poll，But this does not mean that it can rely entirely on the simulation of the big model。First is a problem with data desert，Some places are relatively small online，Not enough data training set，It is difficult to rely on large models to simulate。Followed by data，But may not truly reflect the psychology of the offline people。

Of course，From the long term，Solve data desert is a time problem。can be imagined，If the big model is loaded on the drone to collect data，It can collect terrain and landforms，You can also observe human society，It can also achieve human -machine dialogue。If you can holography、Collect data in real time，Then the current poll data or thesis data is just trace data for it。The key is to install，This is a big problem。

Big models can ask research questions，Can collect data by yourself to verify，This forms a closed loop，You can generate scientific Bet365 app download knowledge。From this sense，Not how to use big models as a tool for future research，Instead, a scientific research process dominated by it。Although more discussions in the near future are talking about human -machine collaboration，But my prediction is that it will become more and more dominant，And people will play a counseling role，Play the role of coach or mentor，Through prompt、Guide to participate in research。As the ability of the big model is getting stronger，It will produce more scientific knowledge，It will also change the method of scientific research。

　Discipline and theory in the new picture 

Qiu Lin: Under the empowerment of the big model，The theory will be from isolated、Single theory，turn into a systematic theory，Become a description of the entire system。Because the previous theory often predicts another variable with one or several variables, it is over。For example，Studies on improving fertility will，I will only consider how to increase the family's fertility willingness by issuing subsidies，End。but actually，If women have three babies，It may reduce the social status of women，Because the time they focus on work are reduced，This will cause women to be unwilling to have children，It will also weaken the effect of the distribution subsidy。So，Study the relationship between two variables or several variables alone，No systematic global view，Is the problem that cannot be really solved or observed。I believe，Big models will change our theory，No longer so isolated，but describing the theory of the entire system。

"China Social Sciences": less and less theoretical in the future，or only one theory? Just like some scholars are discussing the unification of knowledge that ChatGPT may bring? 

Migin: This understanding is still a traditional thinking mode，It will not be explained by traditional concepts in the future。For example, a big platform，If you understand it with traditional economics，It is a monopoly，But the big platform is definitely not a monopoly，Because it is not the concept of traditional enterprises in a physical space，It can be organized in the physical space。For digital space、Digital economy these phenomena，The concept of traditional economics cannot be explained，Now you need to create a new set of interpretation solutions。

"China Social Sciences": What is the situation of human knowledge system in the future，Is it necessary to break the discipline -based construction model? 

Migin: For this question，As Wittgenstein said，Can't say，We should keep silent。

Qiu Lin: After having a big model，We from the original interpretation orientation，We from the original interpretation orientation，Pay more attention to how to solve practical problems。The solution of the actual problems will contain the variables of psychology、variables of economics、Variables of sociology ... It is a solution to the integrated meeting，Is a disciplinary solution。For example, wealth distribution，It is the entire system，The variables are distributed in different fields，The big model produces a fusion interdisciplinary solution。There are many parameters in this solution，Is a very complicated solution，Maybe we can't understand。One of the jobs of a big model，How to tell us the complex theory in the most streamlined language，Become a cognitive、An understanding of a theory。

"China Social Sciences": eliminate disciplines，This is a very big change for the knowledge system。 

Qiu Lin: The reason why theory is important，Let us have a understanding of the world，Provide a framework。Theory is a tool that allows us to live in this world that is not anxious。So，We often pursue simple and easy -to -understand theories。but，This theory is often broad、Not precise、Can't solve actual problems。So，The theory that to be able to truly solve social problems must be complicated、Interdisciplinary。

Gao Qiqi: Disciplinary reconstruction is positive。The current division of disciplines is in the process of the second industrial revolution，Social division of labor spoken by Turkan，It is due to the birth of the new profession，Classification bet365 best casino games of education，Education is actually to cultivate labor。Under this logic，The more divided the more, the more thinner，People have become tools，That is the alienation mentioned by Marx。An important potential of a big model is，It can be an important empowerment tool，No matter what you learn，It seems to help you，I don’t need to be subdivided as it was。From this perspective，It responds to the alienation problem that Marx said。Future，Everyone should be almighty，You can write poems，You can also write code。After a big model,，The entire learning method can be turned into a question and answer。Through encyclopedia，Let individuals establish a global learning ability。

Although the big model cannot be said to be full sample，But it is a very large sample of human knowledge，It can also be said to be a world model，World Model。Knowledge in the world is basically compressed into a big model with knowledge compression，People then call according to the needs。What are its benefits? Since its sample volume is very large，May be infinitely close to reason。When you encounter difficulties，In the past, I could only ask my parents，Ask someone with experience，Maybe earlier can be through divination。Various methods always give you a reason for persuasion。From the perspective of intellectual samples close to full sample volume，The answer given by the big model can indeed make the individual of human beings close to reason。but，To make it persuasive and convinced，It may be a bit troublesome。

This require，About this involve two questions。First is the problem of prediction。The big model is essentially that there is no way to exhaust the possibility，There is no way to predict。If the big model can really get everything，Actually not an open system，but a closed system，It's a bit like setting all the conditions。But this world will become very boring。I think it's still open。The other is a question of cause and effect。I think the cause and effect research of social sciences is basically difficult。Just like the narrative economics，Human society is actually a narrative，Prophecy of self -realization，It seems difficult to exist absolute cause and effect。

Migin: The pursuit of causality in social sciences is too complicated，Multi -cause multi -effect、Multi -caused by different results。The emergence of the emergence is not a causal solution。So modern social sciences put aside cause and effect，Going to the main model -based model、complex system，Concepts of emergence are more suitable for social sciences' views。The emergence of a big model is to establish a plateau generated by the theory，Researchers establish a peak on the new plateau。The relationship between the plateau and the peak，It must be human -machine cooperation。

　　"People" in the era of artificial intelligence 

"China Social Sciences": whether it is the dominant model of the big model，Still human -machine cooperation，New technology obviously plays a very important role in academic research。So，What role does a person plays in it? What are the requirements for this new research model? 

Migin: The emergence of big models，Researchers have richer imagination、Control and deep insights on the overall situation。Under the situation of the popularization of new technologies，The level of the research level depends on the ability of the individual to talk to the big model。The biggest problem now faces is how to regulate。In the past, 25 school alliances in the United States opposed ChatGPT，Recently Harvard University released the ChatGPT Class use guide。Because it realizes that doing homework with ChatGPT is unavoidable。Everyone uses，It is the establishment of a peak on the plateau。

Gao Qiqi: I think the prompt project is a short -term behavior。Because the big model develops quickly，Now with Auto GPT，Those reminder techniques that were originally discussed，The machine can automatically execute。

"China Social Sciences": Is there any disadvantages or attention to using large models for academic research? 

Miginin: It is too late to use now，No disadvantage。

Gao Qiqi: The top international artificial intelligence research expert，For example, the Turing Award winner Bet365 app download Yoshua Bengio and Geoffery Hinton，They are very worried about the development of artificial intelligence。The threat of artificial intelligence is not an illusion。The sample amount of the big model is large enough，It will have the function of correction。But it will produce a greater polarization effect，In fact, there are not many choices in them，It will definitely increase from the public effect。From this perspective，Big models may be the decisive factor that can be elected as the president of the United States。

"China Social Sciences": The answer of ChatGPT may affect the behavior or cognition of the questioner。 

Qiu Lin: In fact，The influence of the media on people has always existed。The two parties of the American people are very serious on social media，It turned out that it was just the voice of some people，But the people in reality are also extreme differentiation。This is actually difficult to say that it is a causal relationship，The polarization on social media affects the real world，It is said that social media reflects the real world。Large model is the same。It reflects people’s thoughts， It will also affect people’s thoughts。

Gao Qiqi: It will become a new battlefield，New public opinion guidance。We want to see the capital and interests behind the big model，The ideology it represents。big models such as ChatGPT can be available，Among them, there is a very important job called "alignment"，Align with the mainstream values ​​of the United States，Otherwise, it will not be available。

Actually，It is for our current researchers，Impact is limited，Because our entire research habits have formed。But for people who have lived under the strong influence of large models since childhood，The challenge is whether he is confident to do his own research? For academic research，The communication of people and people is crucial。If the technology is completely led，Finally, it will inevitably lead to the loss of human subjectivity。Under technical leaders，The communication between people and people will become weaker and weaker，In the end, it will become a communication between adults and technology，Technology will become center。This is actually what Marx said alienation。In this alienation background，Human beings will definitely die，At least decline。Although from Zhuangzi's Qi theory，People and machines are also equal，But humans after discussion after discussion，It is also not in line with the mainstream values ​​of human beings。After all，The machine is invented by people，Even if the result is destined。So why should you accelerate? The most fundamental problem now is，All countries are accelerating，I hope to benefit from it。

　　(This edition is compiled by Liu Qiancai, a reporter from this newspaper)