Posthuman Diagnosis: Evaluating Large Language Models in the Recognition of Celiac Disease

Abstract

As AI systems become integral to clinical practice, their influence on diagnostic knowledge requires critical examination. This study assesses three large language models (LLMs)—ChatGPT-4, Gemini, and AskAi—in diagnosing celiac disease (CeD), a condition often delayed due to its multisystemic complexity. Moving beyond AI as a passive tool, we analyze these LLMs as active epistemic agents within posthuman diagnostic frameworks. Twenty diverse CeD cases were evaluated by each model, intentionally excluding serological/histological data to focus on symptom interpretation. Results show ChatGPT-4 outperformed Gemini and AskAi in accuracy and contextual reasoning, particularly for atypical CeD. However, each model exhibited distinct computational logics, challenging assumptions of AI neutrality and highlighting their unique epistemological biases. This study positions AI as a co-producer of clinical knowledge, advocating for ethical integration, participatory design, and real-world validation in autoimmune diagnostics. By framing diagnosis as a hybrid cognitive practice, it advances equitable and reflexive healthcare paradigms.