Can a generative AI model systematically identify and structure philosophical arguments from classic political texts? In this post, we put OpenAI’s o3 model to the test by applying it to a foundational work in early modern political thought: John Locke’s Second Treatise of Government. Focusing on Chapter II – Of the State of Nature (§4–§15), we instructed the model to extract word-for-word quotations that contain complete philosophical arguments or core normative claims. For each passage, the model was asked to identify the main philosophical question, break the argument down into premises and conclusion, and classify the type of reasoning used (e.g. deductive, normative, historical).
Prompt
The OpenAI o3 model was prompted to perform argument mining on Chapter II (Of the State of Nature, §§4–15) of John Locke’s Second Treatise of Government. The input was provided as a plain-text PDF. The model was asked to extract exact quotations containing full philosophical arguments or core claims, identify the section number, formulate the underlying philosophical question, break the argument into premises and conclusion, and classify the reasoning type. The output was returned in a structured Markdown table format.
You are a political philosophy assistant helping to extract precise arguments from historical texts.
TASK:
From the uploaded PDF (John Locke, Second Treatise of Government), focus only on Chapter II – Of the State of Nature (sections §4–§15).
Identify word-for-word quotations from this chapter that represent complete philosophical arguments or core claims.
For each, extract and return:
- The exact quoted sentence or passage from Locke
- The section number (§) if available
- The main philosophical question the argument addresses
- A breakdown of the argument into premises and conclusion
- The type of reasoning used (see accepted types below)
Output Table Format (Markdown):
- Section (§)
- Quoted Argument
- Question
- Premises
- Conclusion
- Reasoning Type
Recognised Argument Types (Use only these categories):
- Deductive – the conclusion follows necessarily from the premises
- Inductive – the premises support the conclusion probabilistically
- Abductive – inference to the best explanation
- Normative – based on what ought to be done or valued
- Analogical – based on comparison or similarity
- Transcendental – reasoning from conditions of possibility
- Historical – appeals to historical examples or facts
- Reductio ad absurdum – demonstrates contradiction in a contrary claim
Only assign a type if the passage clearly fits one of these categories.
Output
OpenAI o3 approached the task with notable precision and methodological clarity. It demonstrated a structured reading strategy, progressing section by section, identifying core claims, and formulating philosophical questions based on Locke’s reasoning. The model correctly distinguished between normative, deductive, and reductio arguments, while showing awareness of context and conceptual nuance. Its internal monologue reflects a cautious and reflective interpretation process, emphasising the need for verbatim quotation and careful formulation of premises and conclusions.
The extracted output demonstrates a clear and disciplined approach to argument mining. The model successfully identifies core philosophical claims from Locke’s Second Treatise, and reconstructs the associated reasoning with coherence and precision. Each argument is appropriately classified by reasoning type—deductive, normative, analogical, historical, or reductio—showing both textual sensitivity and conceptual understanding. The formulation of philosophical questions is concise and relevant, and the breakdown into premises and conclusions reveals logical structure without oversimplification. While the table does not exhaust every nuance of Locke’s text, it offers a highly usable framework for teaching, commentary, or further analysis.
Recommendations
This experiment highlights the potential of generative AI—specifically OpenAI o3—for structured argument mining in political philosophy. While not a substitute for expert interpretation, the model performs well in identifying and reconstructing key arguments with clarity and philosophical coherence. For researchers, educators, and students, such tools can offer valuable scaffolding in close reading and argument analysis. Future work might explore applying similar prompts to other canonical texts, refining criteria for reasoning types, or integrating human-AI collaboration in philosophical annotation workflows.
The authors used OpenAI o3 [OpenAI (2025) OpenAI o3 (accessed on 1 May 2025), Large language model (LLM), available at: https://openai.com] to generate the output.
