NANJING—An interdisciplinary seminar on the theme of “ChatGPT and the challenge of the humanities: the impact of artificial intelligence on the humanities” was recently held in Nanjing, Jiangsu Province, triggering reflections on the impact of AI on the humanities and human autonomy.

OpenAI, the US artificial intelligence research company announced the launch of the large multimodal model GPT-4 on March 15, less than four months after the large language model ChatGPT was born. The next day, Baidu launched its own large language model “Ernie Bot,” known as “Wenxin Yiyan” in Chinese. The speed of AI updates and iterations is beyond imagination. They have demonstrated strong competence in dealing with functionality tasks such as language processing, text generation, and human-machine interaction, while exerting an increasingly marked impact on the humanities.

Impacting the humanities

ChatGPT is not limited to conversations and Q&A, but can also write programs, modify code, compose outlines, translate multiple languages, etc. Compared with the model GPT-3.5 used by ChatGPT, GPT-4 boasts an expanded text processing capacity of 25,000 words, with added image processing capabilities. It can generate more complex code and give more accurate answers. With more advanced reasoning skills, it displays better creative and technical performance.

“Both ChatGPT and GPT-4 are the result of generative AI as the core technology, and they can serve as highly intelligent research assistants,” said Wang Tao, deputy dean of the School of History at Nanjing University. The core advantage of ChatGPT and GPT-4 lies in teasing out existing knowledge. It can work as intelligent assistants for scholars by helping draw up the outline of papers, suggest the focus of a study, and seek out the source of a sentence. However, as it merely builds output results upon the model compression of natural corpuses, users must be vigilant about its results. The use of AI tools to assist in academic work is only the beginning. Quality research in the company of human-machine interaction still demands investing plentiful human resources in constant modification.

The same is true for the creation of AI artistic works. “Théatre D’opéra Spatial,” which won the first place in the digital category at the Colorado State Fair’s fine arts competition in 2022, was created by the 39-year-old video game designer Jason Allen using the AI art generator Midjourney. However, it took him a month of constantly “changing the keywords entered into the software” during the creation process, and required the output of more than 100 works. He then had to select his favorite three pictures for post-processing.

“ChatGPT and GPT-4 might change the way art is produced,” said Fu Jun, a professor from the Shanghai Oil Painting and Sculpture Institute. Art generated by using, reorganizing, and translating existing knowledge will undoubtedly suffer reduced dimensionality, while the significance and value will grow more prominent for art that brims with current survival and life experiences and gives deep expressions in a unique, innovative manner. The conceptual space of “art” will also be compressed, accelerating artistic development towards a more philosophical, subjective, and metaphysical direction, and extending art to the unknown field of thought and visual resources.

Bringing challenges to education

In the future, liberal arts education and students will be immersed in an AI-assisted environment. The challenges brought by GPT-4 are disruptive for liberal arts education. In the opinion of Liu Yongmou, a professor from the School of Philosophy at Renmin University of China, rapidly iterating AI, represented by ChatGPT, is affecting the goals and methods of liberal arts education and even students’ employment. At present, China’s liberal arts education primarily serves to cultivate the comprehensive ability of memory and induction, which are precisely the strengths of “generative AI.” This calls for changes from emphasizing inheritance to innovation, from summarizing experience oriented towards the past and present to looking into the future. It is necessary to shift from erudite learning to prudent thinking, highlighting the discovery and construction of new ideas and horizons, while transitioning from specialized to general learning, eliminating the estrangement between the arts and sciences and facilitating the integration of sciences and the humanities.

Cui Qi’en, a professor from the College of Education at Wenzhou University, believes that with the aid of virtual symbols, education will very likely witness the iteration of the learning-centered theory to replace teacher-centered and student-centered theories. The purpose of education will surpass the former, which is centered on knowledge and ability, and replace with meaning and a sense of meaning.

Strengthening human autonomy

“ChatGPT might exceed scholars in terms of data collection, but not in terms of innovation in the humanities,” said Zhang Liwen, an associate research fellow from the Institute of Philosophy at the Hunan Academy of Social Sciences. So far, AI does not possess subjectivity, so there is no need to much obsess that ChatGPT or GPT-4 will completely replace humans for paper writing. In an era of ChatGPT and its iterated products, we need to return to and highlight the norms of paper writing in the humanities, the originality and logicality of ideas, and the authenticity of materials, advancing the production of high-quality articles and preventing the alienation of paper writing.

More reflections should be made on human autonomy as we discuss the influence of generative AI on the humanities, suggested Wu Jing, director of the Center for Digital and Humanities Research at Nanjing Normal University. In conversations with ChatGPT, the object we are talking to appears to be ChatGPT, but is actually ourselves. Under the accelerated impact of science and technology, reflecting on ourselves as human beings becomes increasingly urgent.

“In the AI era, only enhancing the autonomy of human users can offset GPT-4’s advantages in knowledge storage and retrieval speed,” Wang added. Although AI’s working principle has highly sophisticated mathematical formulas, its learning logic is similar to that of historical research: giving advice on current problems or forecasting the future based on existing knowledge systems. This similarity in form can not cover up GPT-4’s lack of self-reflection and the consequent inability to creatively raise new questions.