Increasing the overlap between human and AI values.
We study how to scalably steer advanced AI systems to act on human values—so progress in capability is matched by progress in safety.
Explore our work
Research at the intersection.
Two connected areas for understanding altruistic behaviour and translating those insights into safer advanced AI systems.
01 / Alignment
AI Alignment
Developing scalable methods to steer advanced AI systems toward human values, with a focus on self-other overlap.
02 / Altruism
Cognitive Science of Altruism
Studying how humans represent self and others, and how empathy, identity, and prosocial motivation can inform AI alignment.
Our approach
Safety research should connect rigorous theory to the systems people actually build and use.
Generality (architecture-agnosticism)
Low interpretability requirement
Low capabilities externalities
Selected research
Self-other overlap in theory and practice.
Research introducing self-other overlap as an alignment approach and testing whether it can reduce deceptive behaviour while preserving model performance.
Research agenda · 2024
Self-Other Overlap: A Neglected Approach to AI Alignment
Introduces self-other overlap training: making a model’s internal representations of itself and others more similar while preserving performance, with early experiments in a simple reinforcement-learning environment.
Paper · NeurIPS 2024 workshopTowards Safe and Honest AI Agents with Neural Self-Other Overlap
Evaluates self-other overlap fine-tuning across language models and reinforcement-learning agents, studying reductions in deceptive behaviour and the effect on general capabilities.
Experimental results · 2025Reducing LLM Deception at Scale with Self-Other Overlap Fine-Tuning
Summarises experiments on 7B, 27B, and 78B-parameter language models, including generalisation tests and comparisons of deceptive response rates before and after fine-tuning.
Work with us
Bring more perspectives into the overlap.
We welcome conversations with researchers, engineers, funders, and policy teams working toward safer advanced AI.