How to Teach Artificial Intelligence to Speak Human Language? 》Xu YingjinThe Commercial Press, 2023SORA and ChatGPT have repeatedly refreshed the imagination of human beings about AI
But in the view of philosophy professor Xu Yingjin, ChatGPT is not really artificial intelligence.
On the occasion of the launch of his new book, let's hear how he answers 8 questions about AI
Q:
What is ChatGPT, which has caused heated discussions? Xu:The number one event in the international artificial intelligence community from 2022 to 2023 is the birth of ChatGPT developed by the American Open AI company. In the industry, software that is more powerful than the traditional "chatbox" is generally called a "chatbot". Whereas, ChatGPT falls under the category of "chatbots". However, from the perspective of the development history of artificial intelligence, such platforms are a combination of "chatbox" technology and deep learning technology that have appeared for a long time, and do not reflect real innovation at the scientific level. The future that ChatGPT reveals to us is likely to be an era of philistinism with smooth grammar, an era in which even people's self-deceptive instincts may degenerate.
This is the answer provided by ChatGPT after all, how can it be wrong? ”
The French philosopher Sartre put forward the theory of "self-deception".
Q:Is the artificial intelligence you just mentioned the humanoid robot in science fiction film and television works? Xu:AI and robotics are two different disciplines, let alone humanoid robot research. Human beings actually have a tendency to "animist" psychological projection, that is, many non-living beings with animal or human forms are regarded as souls. As long as the main creator of the film and television designs the relevant robot to look like a human, such visual input will successfully stimulate the audience's personification tendency, so as to autonomously endow such a robot with wisdom and soul. But such a flattering approach simply doesn't work in actual AI research. Specifically, to give an actual AI body any kind of actual operational function requires a tremendous amount of effort from the programmer in the background. Although AI is an academic field with a high degree of technical integration, its commercial application is very wide. Due to the information asymmetry between the two, there is often a huge gap between the "internal image" of AI in the professional world and its "external image" in the eyes of the public, including political and business elites. 
Sophia, a humanoid robot (illustrated in the book).
Q:Is it possible for AI to be like Moss in "The Wandering Earth", sacrificing individual humans in the name of "protecting humanity"? Xu:That's not to worry too much about. From the perspective of dramatic conflict, the overall cold perspective of AI in these film and television works is in stark contrast to the local (but at the same time more warm) perspective of individual human beings. AI lacks emotion, but it surpasses humans in terms of computing power. As a result, AI can better understand what the "big picture" is than humans – even if it's not the "big picture" that individual humans want to accept. But the above impression is based on a deep misunderstanding of AI, because superior computing power does not mean that you have control over the whole world. In fact, for any agent to grasp such global knowledge, it also needs a super knowledge graph for all problem domains, and this knowledge graph is often the crystallization of human intelligence.
As we all know, we humans do not yet have a general consensus on the shortness and length of these positions. In other words, there is no unified conceptual premise for solving the "tram problem", let alone building a unified knowledge graph under this premise. From this point of view, as a form of human intelligence transfer, no AI can go beyond the current upper limit of human intelligence and give a flawless knowledge graph for the solution of all problems.
MOSS model (**edited by Yu Xiongfei).
Q:Do you mean that science fiction films and television works mislead everyone's imagination of AI, so in reality, do we have a specific coordinate for measuring the degree of development of artificial intelligence? Xu:Yes. Due to the dominant position of film and television dramas in the modern transmission system, mainstream science fiction movies with AI as the theme have shaped the public's impression of AI to a considerable extent, and caused the misinterpretation of AI by relevant film and television creators to spill over, leading to misinterpretation by the whole society.
The author's practice is the Anglo-American analytic philosophy of language, so after I began to pay attention to the philosophy of artificial intelligence around 2006, I began to systematically focus on natural language processing in the field of artificial intelligence. How to Teach Artificial Intelligence to Speak Human Language? is a collection of the author's thinking results in related fields.
The so-called "artificial intelligence" is a discipline that uses the technical means provided by computers to simulate or partially simulate human intelligence. And a very important aspect of human intelligence is "talking". If "being able to speak" is one of the most important metrics that distinguish human intelligence from animal intelligence, then AI in its entirety should clearly have this characteristic as well.
How to Teach Artificial Intelligence to Speak Human Language? Inner page.
Q:So has the current AI reached the point of "talking"? Xu:First of all, we would like to introduce you to a concept:
In the AI world, computer programming research that is responsible for completing tasks such as email processing, automatic generation of reading summaries, automatic translation, and automatic text generation is labeled as "natural language processing" (NLP).
The various bot chatboxes are specific NLP mechanisms designed for different NLP tasks, rather than a blanket solution to all NLP problems.
Some language materials that are difficult to obtain through big data technology in principle are also difficult to fully process with the current mainstream NLP technology. These materials include: (a) rhetorical devices such as "puns", metaphors, and irony in human language. (b) Weak languages, such as minority languages and local dialects, that lack sufficient online data samples.
Q:The recent launch of SORA has raised a new round of concerns about the rapid development of AI models. What can philosophy tell us at this time? Xu:The task of the philosophical researcher is to expose the unstated premises of NLP research and to reflexively critique them. At present, the topic orientation of NLP academic research is completely dominated by accidental engineering needs, and lacks a holistic approach at the philosophical (or even scientific) level. What's more, to the best of the author's knowledge, the philosophical community of language currently lacks sufficient theoretical impulse to fully engage in the study of NLP. This state of "two skins not disturbing each other" is obviously unsatisfactory.
In general, the relationship between philosophy and NLP research is not fundamentally different from the relationship between philosophy and science and engineering research planning in general. That is to say, as with many natural science research projects, different NLP research projects already presuppose philosophical positions of one kind or another, but relevant empirical researchers are often not interested in reflecting on such positions. Since the study of NLP will have to presuppose one or another view of the nature of language, the relationship between philosophy of language as a branch of philosophy and the study of NLP becomes unusually close.
SORA official webpage.
Q:How to Teach Artificial Intelligence to Speak Human Language? What are the characteristics of this book compared to other popular books related to AI in the book market? Xu:First, this book is definitely not a "pie-in-the-pie" AI science book, that is, it promises readers a beautiful and empty future of AI development, so as to secretly help some technology investment solutions that are not so mature.
Second, this book focuses on the NLP problem in AI research, and attempts to reflect on the philosophical and theoretical issues behind the NLP problem from a philosophical perspective.
Third, in addition to the reliance on philosophical resources, the book will also contain a considerable amount of discussion of linguistic issues, especially various analyses based on cognitive linguistics. At the same time, the discussion of small language corpora such as Japanese will also occupy a relatively high weight in the discussion of this book. Such a discussion will not only enrich the empirical material covered in this book, but also help readers understand the diversity and complexity of human language phenomena. To a large extent, this approach echoes the aforementioned vigilance about the hegemony of English thinking.
How to Teach Artificial Intelligence to Speak Human Language? Back cover.
Q:As a cross-border scholar who is both a scholar and a philosopher, did you make any interesting discoveries while writing this book? Xu:The passage that the author translated with ChatGPT comes from the corpus in the history of the Three Kingdoms written by the author himself. We know that it is relatively easy to systematically ** "drink" without "rice" appearing directly (the answer is, of course, "no"), but if you want to ** whether "rice" will appear after the "drink" interval of fifteen words, it is another matter. In the context of big data technology, proponents of the statistical approach respond to Chomsky's criticism in the following way: as long as there is enough data, we can "control the complexity" with the help of the aforementioned "word embedding" and "attention" mechanisms—that is, we can determine the affinity between distant words through more data comparisons. However, the above technical path is difficult to cope with NLP processing tasks in unpopular knowledge fields (for example, when the author tried to use ChatGPT to translate the romanized name of Japanese philosopher Nishida Kitaro "Nishida Kitarō" into Chinese, the system unexpectedly output the answer of "Sakamoto Ryoma"). Author: Authentic "slash" middle-aged
Born in 1978, he is a native of Shanghai. He is a professor at the School of Philosophy of Fudan University and a consultant of Pujiang National Laboratory, specializing in Anglo-American analytic philosophy and artificial intelligence philosophy. He is the author of "Fifteen Lectures on the Philosophy of Artificial Intelligence", "Mind, Language and Machines", "Philosophy of Use", etc. In his spare time, he likes to read Chinese and foreign history, appreciate films and television from various countries, and create a five-volume long history ** "Jian: The Prequel of the Three Kingdoms: Sun Jian Kuanghan".
In addition to her main business, Xu Yingjin is also a hardcore military model player, an amateur soul painter, and an "Internet celebrity" philosophy blogger who is active in Douyin, Xiaohongshu, and Bilibili, and has gained countless fans (account: Xu Yingjin Wen Shi Zhe).
Hand-painted work by Xu Yingjin.
Let's read "How to Teach Artificial Intelligence to Speak Human Words? "This book! 
Fudan General Education Library.
Fudan UniversitywithThe Commercial PressThe large-scale general education series jointly launched by the company covers the themes of literary and historical classics and cultural inheritance, philosophical wisdom and critical thinking, dialogue among civilizations and world vision, scientific and technological development and scientific spirit, ecological environment and life care, artistic creation and aesthetic experience, social research and contemporary China, etc., and is published 3-5 kinds every year. Bibliography coming soonXu Yingjin "How to Teach Artificial Intelligence to Speak Human Words? Ren Junfeng, The Rise and Fall of Empires: The Political World of Thucydides, Guo Xiaodong, The Historical World of Spring and Autumn and Confucius, and Wu Xiaoqun, The Tragic World of the Ancient Greeks, ......