AI ethics in action.
Guided by our AI Ethics principles of accountability, responsibility, and transparency, we’ve created standardized processes from design to development to deployment, including training, testing, and a review process overseen by a diverse AI Ethics Review Board.
Training
AI is only as good as the data on which it’s trained, and what results are considered appropriate depends on each use case. That’s why we build datasets specifically to meet the needs of each of our businesses, to ensure we have diverse and ethical results that are appropriate for the way in which the AI will be used.
Testing
We conduct rigorous and continuous testing of AI-powered features and products to mitigate against harmful biases and stereotypes. This includes automated testing and human evaluation.
Impact Assessments
Engineers developing any AI-powered features submit an AI Ethics Impact Assessment. This is a multipart impact assessment designed to identify features and products that can perpetuate harmful biases and stereotypes. This allows our AI Ethics team to focus their efforts on features and products with the highest potential ethical impact, without slowing down the pace of innovation.
Diverse human oversight
AI-powered features with the highest potential ethical impact are reviewed by a diverse cross-functional AI Ethics Review Board. Diversity of personal and professional backgrounds and experiences is critical to identifying potential issues from a variety of perspectives that a less diverse team might not see. Hear from members of our board. Watch the video.
Feedback
We provide feedback mechanisms so users can report potentially biased outputs and we can remediate any concerns. AI is an ongoing journey, and we want to work together with our community to continue to make our tools and products better for everyone. If you have a question about AI ethics or want to report a possible AI ethics issue, please contact us.