Abstract:

This paper focuses on the application of large language models in psychobiographical research, systematically examining the efficiency revolution they bring and the profound crisis of researcher subjectivity they may induce, while innovatively proposing a comprehensive framework to address these challenges. Its core contribution lies in moving beyond the simplistic debate over “whether to use AI” and constructing a theoretical research framework for “how to facilitate human-machine collaboration.”

The innovative aspects of this paper are mainly reflected in the following three areas:

1) Systematic and dialectical deconstruction of the application potential and inherent limitations of LLMs throughout the psychobiographical research process.For the first time, using the “puzzling question” analytical model in psychobiography as a framework, the paper meticulously analyzes the specific roles and boundaries of LLMs at each stage of the research process. Through case studies, it demonstrates how LLMs leverage their massive data processing and pattern recognition capabilities to significantly enhance research efficiency and expand researchers’ informational horizons in steps such as “rapidly identifying potential subjects,” “assisting in formulating puzzling questions,” and “collaboratively optimizing text.”

However, the paper points out that behind these efficiency advantages lie fundamental epistemological limitations. The operation of LLMs is based on statistical probability rather than genuine understanding, which creates an insurmountable gap in the core aspects of psychobiography. For example, in the stage of “formulating puzzling questions,” LLMs can only generate lists of factual “irrationalities” but cannot achieve the creative leap from “behavioral anomalies” to “meaningful puzzles”—a leap that relies on the researcher's lived experience and existential engagement. This limitation stems from LLMs’ inherent constraints in semantic comprehension and meaning generation, which is also the root cause of superficial risks such as data bias and false content generation.

2) Explicitly propose and deeply elaborate on the core challenge of the “crisis of researcher subjectivity”, revealing its progressive manifestations.

Building on an analysis of various technological risks, this paper suggests that the widespread application of LLMs may lead to the erosion and relinquishment of the researcher’s dominant role in meaning interpretation and theoretical creation. Researchers may progress from merely enjoying the efficiency of LLMs in data processing to developing dependence on them for core intellectual activities such as question formulation and theoretical framework selection, thereby outsourcing the creativity that should inherently belong to the researcher to algorithms.

When confronted with the structured, professionally articulated analyses produced by LLMs, researchers may unconsciously suppress their own critical thinking, defaulting to the algorithm's “optimal solution,” which could result in a loss of the critical and creative essence essential to the humanities. The paper clearly states that addressing this crisis requires more than mere technical improvements; it demands systematic reconstruction across three levels: data foundation, ethical norms, and methodological processes.

3) Constructing a researcher-led framework to address the crisis.

This paper proposes the development of specialized databases based on rich local historical materials (such as Chinese classical texts) and the training of domain-specific LLMs using these resources, thereby establishing a virtuous cycle of “local database——training specialized models——supporting local research——feedback for model correction.” This approach aims to ensure the accuracy and sensitivity of cultural interpretation from the data source, overcoming the cultural blind spots and biases of general-purpose LLMs. At the same time, the paper proposed ethical principles including Human Agency and Accountability, Fairness and Bias Correction, Privacy Protection and Data Security, and Risk Prevention and Dynamic Governance. These principles establish clear behavioral boundaries for responsible technology use.

Finally, the paper designs and elaborates in detail an iterative, researcher-led human-computer collaborative research process. The core feature of this process is “the researcher defines the framework, the LLM executes the computations.” Key steps include: manual verification and purification of materials by the researcher; transforming psychobiographical theories into specific analytical dimensions and annotation rules, which are then “injected” into the LLM through fine-tuning or prompt engineering; the researcher ultimately determining the puzzling question based on resonance with the materials; and the researcher reviewing the theoretical fit of the LLM’s coding results. Throughout this process, the LLM consistently serve as “efficient assistants” for processing information, providing options, and generating drafts, while the power to ask questions, establish rules, make critical judgments, conduct in-depth interpretations, and construct theories remains firmly in the hands of researchers.

In summary, the novelty of this work lies in providing psychobiography, and qualitative research more broadly, with a comprehensive theoretical framework and practical guide for effectively utilizing technological tools in the AI era while steadfastly upholding the core values of deep hermeneutic interpretation and theoretical innovation inherent to humanities scholarship.

Key words: artificial intelligence, psychobiography, large language models, human-machine collaborative, DeepSeek