Our mission

Empowering discovery. Respecting human judgement.

We develop methods for knowledge discovery based on transparent and trustworthy AI. We believe that AI should augment human reasoning in the pursuit of knowledge, not replace it.

Our research:

• [January 2026, paper] – We introduce Semantic Pullbacks, a theoretically grounded explanation method that achieves substantial qualitative and quantitative improvements over existing approaches, while enabling the generation of compelling counterfactual examples at low computational cost. Semantic Pullbacks generalise Excitation Pullbacks to Transformer architectures.
• [July 2025, paper, code, demo] – We propose Excitation Pullbacks, a novel way to explain AI models. Our method shows what the model really looks at when making a prediction. It amplifies the most important features for the chosen label, and these features turn out to align surprisingly well with human perception. This makes AI decisions easier to understand and more transparent.
• [March 2024, paper, code] – We introduce semantic features as a candidate building block for transparent neural networks and build fully explainable and adversarially robust PoC neural network for MNIST.