Beyond the Black Box: Identifiable Interpretation and Control in Generative Models via Causal Minimality
This addresses the need for transparent and reliable AI systems by offering a principled foundation for interpretability in generative models, though it builds incrementally on existing methods like sparse autoencoders.
The paper tackled the problem of deep generative models being opaque black boxes by using the principle of causal minimality to provide theoretical guarantees for interpretable latent representations, enabling clear causal interpretation and robust control in diffusion vision and autoregressive language models, with empirical results showing extraction of hierarchical concept graphs and fine-grained model steering.
Deep generative models, while revolutionizing fields like image and text generation, largely operate as opaque black boxes, hindering human understanding, control, and alignment. While methods like sparse autoencoders (SAEs) show remarkable empirical success, they often lack theoretical guarantees, risking subjective insights. Our primary objective is to establish a principled foundation for interpretable generative models. We demonstrate that the principle of causal minimality -- favoring the simplest causal explanation -- can endow the latent representations of diffusion vision and autoregressive language models with clear causal interpretation and robust, component-wise identifiable control. We introduce a novel theoretical framework for hierarchical selection models, where higher-level concepts emerge from the constrained composition of lower-level variables, better capturing the complex dependencies in data generation. Under theoretically derived minimality conditions (manifesting as sparsity or compression constraints), we show that learned representations can be equivalent to the true latent variables of the data-generating process. Empirically, applying these constraints to leading generative models allows us to extract their innate hierarchical concept graphs, offering fresh insights into their internal knowledge organization. Furthermore, these causally grounded concepts serve as levers for fine-grained model steering, paving the way for transparent, reliable systems.