Our approach to safe and ethical AI
We’ve seen first-hand the opportunity AI can provide to tackle real business challenges - but it can raise problems, too.
AI ethics
Our ethical AI principles
AI Safety
Fair
Robust
Explainable
Private
Asymmetric Shapley values: incorporating causal knowledge into model-agnostic explainability
A mathematically justified method for incorporating causality into Shapley values, overcoming a fundamental limitation of current day explainability. Published in NeurIPS.
Shapley explainability on the data manifold
A demonstration of how to calculate Shapley values, the most commonly used method to explain black-box models, without making the flawed assumption that features in the data are independent, developing the first general, scalable approach to correctly calculating Shapley values. Published in ICLR.
Human-interpretable model explainability on high-dimensional data
A framework for explaining black-box models in terms of the high-level features (e.g. people’s characteristics, objects) that humans care about, rather than the low-level features (e.g. pixels) that the models use.
Explainability for fair machine learning
The first demonstration of directly explaining bias in AI models. Submitted.
Learning to Noise: Application-Agnostic Data Sharing with Local Differential Privacy
A method to share high-dimensional data with the powerful guarantees of Local Differential Privacy. Submitted.
Our approach to building ethical, safe AI systems
We ensure AI Safety is embedded within every stage of the development lifecycle.
Our data science teams implement explainability and robustness directly into the model to help users understand what drives predictions.
Built-in explainability fulfils regulatory requirements and flags biased models. We also ensure the system is robust to produce high-quality outcomes.
Users receive explanations and a credibility score with every prediction made, so teams can always trust predictions and confidently action them.
Our teams maintain the model by monitoring explainability, robustness, potential bias as well as pre-empting data drifts before they impact prediction quality.
