Generative AI for Education, Research and Discovery: Issues of Conjectures and Refutations

This is the title of a famous essay by Heinrich von Kleist most probably written in 1805/1806 when Kleist was living in Knigsberg [1].It has the form of a letter from Kleist to his friend Otto August Rhle von Lilienstern.His advice is: "If you want to know something and cannot find it through meditation, [...] talk about it with the next acquaintance who comes across it."Nowadays, Large Language Models (LLM) and Generative Pretrained Transformers (GPT) such as ChatGPT come into play serving as conversationalists.There is even a recent debate about co-authorship of generative AI (GAI) on scientific publications [2] and [3].The gradual fabrication of thoughts is a core phenomenon of research, of discovery, and in education.In a letter to Karl Popper of September 11, 1935, Albert Einstein put it like that: "I think (like you, by the way), that theory cannot be fabricated out of the results of observation, but that it can only be invented."(Translation in Appendix xii of the 1995 reprint of [4] by Routledge, p. 458).That invention is what conversationalists may support substantially, just as Heinrich von Kleist put it.The authors have some doubts that the claim describing "the public version of ChatGPT [...] perfectly responding to any human requests described in natural language" as put by (Wu et al.,[5], p. 1122, reflects the reality appropriately.A considerable number of reports on the successful usage of ChatGPT and its impact on students are clearly glossed over.The deployment of LLMs and GPTs in science and higher education needs systematic efforts [6] including the understanding that system responses are error-prone and hypothetical."ChatGPT requires strong critical thinking" [7].It is not helpful neither in science nor in education, if enthusiasm turns into blind faith. Initial SituationAI is considered decisive to the future of employment [8] as discussed in some detail in [9], [10], [11], and [12].This implies a wide audience's interest in an understanding of the state of affair.The task is bedeviled by publications with euphonious titles such as [13], this one in its section captioned "What is ChatGPT?", pp.2-4, not at all explaining what ChatGPT is, but delusively promising ChatGPT's "ability to understand language input" Journal of Contemporary Education Theory & Artificial Intelligence