Education and the Epistemological Crisis in the Age of ChatGPT

ABSTRACT

The emergence of ChatGPT and its ability to imitate student work has precipitated a crisis in education. Responses range from practical efforts to detect or prevent ChatGPT use to existential reflection about learning and assessment. However, the crisis is also epistemological, with educators struggling to determine what students know and fearing that student knowledge is diminishing. The educational philosophies of Socrates, Dewey, and Freire suggest that recognition of ignorance of different kinds (inquisitive, predictive, and practical) on the part of educators is uncomfortable but essential for improving the aims and methods of education. In this way, ChatGPT and the philosophy of education can be mutually illuminating.

Keywords:

• ChatGPT
• AI
• philosophy of education
• educational crisis
• plagiarism
• ungrading
• Plato
• Socrates
• Dewey
• Freire

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 OpenAI has released several models of GPT. I am mostly referring to 3.5, the model which was released in November 2022 and is free to use as of this writing. ChatGPT is not the only natural language chatbot available to the public, but it is the one that has received the most attention and broken records as the fastest-growing consumer application in history. Due to its popularity, ChatGPT will serve in this article as a stand-in for any natural language chatbot with similar capacities.

2 Older chatbots, going back to the 1960s, had been programmed with rules for responding to inputs. Such chatbots could only respond about specific topics or serve specific functions, such as a therapist or a travel assistant. One of the earliest chatbots that sparked curiosity was ELIZA in 1964-66, which was programmed with a “script” based on a Rogerian psychotherapist. According to the designer, the role of psychotherapist was chosen precisely because due to conventions it is common for the psychotherapist to follow up whatever the patient says with questions and such questions are not interpreted to mean that the psychotherapist has a small knowledge base, which was indeed the case with ELIZA (Weizenbaum 1966).

3 Another ChatGPT feature which might seem to demonstrate learning is that it can remember inputs or commands and adapt its responses accordingly. For example, it can take on the role or “persona” of a famous person or even follow a “flipped interaction” pattern where it is the one asking questions and evaluating responses (White et al. 2023). Such patterns, however, are only remembered during a particular conversation and do not modify ChatGPT’s overall knowledge base and response patterns.