The rapid advancement of edge intelligence in Industrial Internet of Things (IIoT) is transforming human–computer interaction from conventional “command execution” to complex “human–AI deep collaboration”. Within such safety-critical industrial environments, establishing robust mutual understanding and trust mechanisms becomes a significant prerequisite for decision reliability and efficiency. However, existing industrial interaction systems predominantly focus on task progression and explicit command responses, lacking fine-grained, dynamic tracking of operators’ trust states, cognitive evolution, and behavioral dynamics. Moreover, current LLM-based user simulation in evaluation often exhibit an “over-cooperation” bias, failing to capture the cognitive conflicts and trust crises characteristic of high-pressure, high-risk industrial conditions. To address these challenges, we first propose a trust-aware user behavior model, which utilizes an LLM-parameterized Hidden Markov Model (HMM) to formalize collaborative trust as a dynamic latent variable, thereby structurally characterizing the psychological and behavioral dynamics of operators across multi-turn interactions. Building on this, we introduce TATA, a task-oriented agent framework integrating trust-awareness and cognitive alignment. Through a dual-track state monitoring mechanism and adaptive interaction policy coordination, TATA effectively advances collaborative tasks and fosters relationship maintenance in realistic collaborative environments. Comprehensive evaluations on six industrial task scenarios demonstrates that TATA achieves an optimal balance between collaboration depth and task efficiency, outperforming the strongest baseline by achieving 1.6 to 2.6 times higher collaboration efficiency and an absolute increase of over 15 percentage points in task completion rate. These findings provide valuable insights for developing resilient and adaptive deep human-AI collaboration tailored to IIoT scenarios.

Primary light chain amyloidosis is a rare hematologic disease with multi-organ involvement. Nearly one-third of patients with amyloidosis experience five or more consultations before diagnosis, which may lead to a poor prognosis due to delayed diagnosis. Early risk prediction based on artificial intelligence is valuable for clinical diagnosis and treatment of amyloidosis. For this disease, we propose an Evolutionary Neural Architecture Searching (ENAS) based risk prediction model, which achieves high-precision early risk prediction using physical examination data as a reference factor. To further enhance the value of clinic application, we designed a natural language-based interpretable system around the NAS-assisted risk prediction model for amyloidosis, which utilizes a large language model and Retrieval-Augmented Generation (RAG) to achieve further interpretation of the predicted conclusions. We also propose a document-based global semantic slicing approach in RAG to achieve more accurate slicing and improve the professionalism of the generated interpretations. Tests and implementation show that the proposed risk prediction model can be effectively used for early screening of amyloidosis and that the interpretation method based on the large language model and RAG can effectively provide professional interpretation of predicted results, which provides an effective method and means for the clinical applications of AI.