A Study on the Linguistic Features and Brain Cognitive Response Differences of Human and AI Texts

After the rapid development of generative large language models, the boundary involved in human and AI texts has become increasingly blurred, and academic integrity and intellectual property issues therefore need to be reconsidered. At present, many studies focus on automatic identification methods from the perspective of surface linguistic features, while they give limited explanation of the deep cognitive and neural phenomena involved when humans process the two types of texts. The present study combines computational linguistics, behavioral experiments, and neurocognition for examination, and makes a detailed analysis with electroencephalography (EEG) and eye-movement synchronization as the basic approach. The paper first summarizes the stable differences between human and AI texts in syntax, semantics, pragmatics, and other aspects, then discusses behavioral or eye-movement performance according to processing depth. It then interprets the N400, P600, and late positive component observed in fixation-related potentials as neural markers, and finally considers the relationship between factors such as genre and current research. The paper analyzes the key problem of the absence of whole-chain evidence for “linguistic features - neural responses,” and also proposes future directions such as sample expansion and cognition-guided detection. The framework proposed here has theoretical significance for AI text identification and the naturalness of large models.