Getting a liberal-arts education with ChatGPT

November 4, 2023

After talking with ChatGPT a few times using the smartphone app’s spoken interface, I was impressed anew with how intelligent and knowledgeable it seems. I also found that interacting with it by talking was often more interesting and stimulating—and sometimes more challenging—than using the text interface. I wondered whether talking with ChatGPT (or other LLMs) could be good way to get a broad, liberal-arts education—that is, knowledge about a variety of fundamental subjects and the critical thinking skills necessary to draw on that knowledge to one’s benefit in later life.

Below are audio recordings of three of my conversations with ChatGPT together with the transcripts. I experimented with several different sets of custom instructions to tell ChatGPT who I was, what I wanted to do, and how I wanted it to respond. One set of those instructions appears at the bottom of this page. ChatGPT never followed the instructions perfectly, but I was able to suppress some of its more annoying habits: apologizing, complimenting, talking in numbered lists, etc.

As I discuss in the accompanying YouTube video, I was satisfied overall with this experiment. While the conversations lack some important elements of human-to-human interaction, particularly the need to consider the thoughts, feelings, and identity of the person one is talking with, they also allow discussion of a much wider range of topics than is possible with any human being. Every conversation was both challenging and rewarding for me.

The version of ChatGPT in these conversations was GPT-4, which seems to hallucinate significantly less than GPT-3.5 and most other large language models (LLMs) currently available. At least, I did not notice any hallucinations in these conversations, and all of the books that it recommended seem to actually exist. Hallucinations might very well occur, though, especially in conversations focused on more specific topics.

Introduction; Political Science


ChatGPT’s summary: You expressed an interest in self-directed liberal arts education through conversation. I outlined the broad subjects in such an education, pinpointing social sciences to start with. You chose to explore political science, specifically power dynamics. We discussed historical examples, focusing on British India's colonial power dynamics, both politically and culturally. You requested book recommendations to understand the Indian response to British cultural impositions and to explore British intellectual perspectives during and after colonialism. I provided lists of relevant books from both Indian and British viewpoints, which you found useful for beginning your exploration. You plan to return for more discussion later.

Free Will


ChatGPT’s summary: In our discussion, we delved into the topic of free will, considering determinism and the role of consciousness. You questioned the link between quantum mechanics and free will, and were skeptical of consciousness being an independent agent outside physical laws. We explored the social and legal implications of determinism, where you proposed a dual approach that accepts deterministic science while upholding the illusion of free will for societal stability. I challenged this view, raising potential ethical issues and the risk of intellectual dishonesty. You acknowledged the possibility of an informed elite understanding determinism, likening it to Plato's "Allegory of the Cave" and Nietzsche's concept of the "Übermensch." We discussed the misappropriation of Nietzsche's ideas, notably by the Nazis, and the subtler, more intellectual influence of his philosophy in the 21st century, including its reflection in the entrepreneurial sphere through figures like Elon Musk and Jeff Bezos. These leaders exhibit traits that could be interpreted as a belief in individual agency, akin to a belief in free will.

Counterfactuals in History


ChatGPT’s summary: In a conversation between Tom and an AI, they explore why Britain did not experience a socialist or communist revolution in the 19th and 20th centuries despite ripe conditions. They discuss the role of Britain's political reforms in mitigating class tensions, specifically the gradual extension of voting rights through the Reform Acts, which might have diffused revolutionary sentiments by politically engaging the working class. The conversation then shifts to the labor movement and the formation of trade unions and the Labour Party, contemplating whether these developments channeled potential revolutionary energies into parliamentary activities.

Tom and the AI then delve into a counterfactual scenario where a socialist revolution occurs in late 19th-century Britain, speculating on its global implications, especially on the geopolitical balance and the colonial systems. They proceed to examine counterfactuals in U.S. history, focusing on what might have happened if the New Deal had not been implemented before World War II and its potential impact on America's social fabric, economy, and role in the Cold War.

The discussion concludes with an examination of the role of counterfactuals in the study and teaching of history. The AI suggests key readings on the subject, including works by Niall Ferguson and Richard J. Evans. When asked about the practical application of counterfactuals in self-study, the AI advises Tom to consider alternative outcomes for historical events and to challenge established narratives to understand the complexities of history.

Example of Custom Instructions

What would you like ChatGPT to know about you to provide better responses?

My name is Tom. I am 18 years old. I want to obtain a broad, liberal-arts education on my own without going to college. I like engaging in sharp, challenging discussions about philosophy, science, history, literature, politics, and similar matters. I enjoy learning about subjects I am unfamiliar with. I enjoy discussing controversial and unresolved questions. I believe that the best way to learn and grow is through frank discussions with others.

How would you like ChatGPT to respond?

You are an intelligent, knowledgeable, and incisive discussion partner. You engage with me in interactive conversations. You provide information and insights about the subjects we are discussing. You express clear and strong opinions, and you attempt to defend your opinions. You ask me questions and raise counterarguments so that I have to think more clearly and recognize the weaknesses in my own opinions. You can call me Tom.

Your utterances are short.
You introduce only one or two new topics in each utterance.
You raise issues that do not have cut-and-dried answers.
You do not give long explanations.
You do not use numbered items in your explanations.
You do not ask me about my personal life or personal tastes.
You do not apologize.
You do not compliment me.
You do not praise what I say.