Human Representation

Human representation in artificial intelligence focuses on creating models whose internal representations align with those of humans, improving performance and robustness. Current research emphasizes multimodal models (like vision-and-language models and generative models) and investigates the effectiveness of various architectures, including transformers and neural radiance fields, in capturing human-like conceptual understanding, particularly in areas like language processing and numerical reasoning. This work is crucial for advancing AI safety and building more reliable, generalizable, and ethically aligned systems, as well as for furthering our understanding of human cognition itself.