<p>This paper introduces a human–machine hybrid intelligence approach grounded in causal representation, designed to address prediction and decision-making challenges under non-independent and identically distributed (non-i.i.d.) data. The proposed method integrates causal inference with machine learning techniques. Specifically, a Causal Effect Variational Autoencoder (CE-VAE) is employed to extract causal relationships from the data. Subsequently, a Bayesian Neural Network is utilized to fuse the outputs of human and machine intelligence, enhancing the system’s robustness and decision-making capabilities in complex environments. The model framework is structured into three hierarchical levels. The first level involves the independent processing of information by both machine intelligence and human intelligence components. At the second level, features are transformed into latent causal variables using the CE-VAE. The third level then fuses these latent variables via the Bayesian Neural Network, generating a joint causal representation and producing the final decision. Experimental results demonstrate the efficacy of this approach in an unmanned aerial vehicle (UAV) combat intention assessment task. The proposed method significantly improves the model’s accuracy, interpretability, and stability. This research offers a novel perspective on tackling non-i.i.d. problems and holds potential application value in areas such as military decision-making.</p>