Generative AI for brain-computer interfaces decoding: Advances, challenges and future

Brain-computer interfaces (BCIs) establish direct communication pathways between the human brain and external devices, enabling novel modes of human-machine interaction and providing essential technological foundations for innovative clinical therapies. BCI decoding can be divided into discriminative and generative approaches. Discriminative decoding models are primarily used for predefined tasks such as emotion recognition, limited performance in reconstructing high-dimensional semantic content such as natural language or visual images. Recent advances in generative artificial intelligence (AI), including diffusion models and autoregressive Transformers, have brought generative decoding to the forefront of BCI research by enabling the reconstruction of semantically rich content from neural activity. Current studies further demonstrate that decoding visually evoked neural activity into images can achieve structural similarity index measure (SSIM) scores ranging approximately from 0.26 to 0.43. Furthermore, generative AI enhances BCI decoding through data augmentation and sensor optimization, thereby improving both performance and generalizability. Here, we review recent progress in generative AI-enabled BCI decoding, focusing on methodological advances in language and visual decoding. We also summarize its roles in data augmentation and sensor optimization, outline key challenges in decoding accuracy, cross-subject generalization and fairness, and suggest future directions such as multimodal integration and fairness-aware frameworks.