Enabling fair and responsible AI starts with the ability to mitigate the challenge of bias in AI models. We have published technical tutorials on bias as well as guidance for organizations to address bias.

AI safety is central to our work. People need to trust that AI models are safe, compliant and robust. Ensuring accountability and reliability in AI is critical to the successful application of AI models, and putting processes in place for model validation — alongside other model governance processes ensures that algorithms do what they’re supposed to do and don’t do what they’re not supposed to do. We develop new validation methods and we invest in research to deal with the scale, complexity, and dynamism of AI.

AI poses unique data privacy challenges. Addressing data privacy, while leveraging large data sets is an essential component of responsible AI and is critical to responsible AI adoption. We focus on differential privacy in our research, and built Private Data Generation™ toolbox to offer synthetic ML data samples, a method that allows scientists to use large data sets without risking the exposure of personal identifiable information. This tool can be used by researchers to advance the field of AI privacy by proposing novel solutions to this critical issue. We have also published tutorials on differential privacy, among other topics.

Explainability is a key component to trust in machine learning. An understanding of how the ML algorithm learns and makes decisions. Our research, tutorials and interviews with practitioners in the field outline how explainability can help build trust in AI.