**Making Our Products More Inclusive: The Challenge of Representative Data**
**Retraining Machine Learning Models with Inclusive Datasets**
In our commitment to inclusivity, we have encountered various challenges in making our products more inclusive. One significant challenge we faced was finding and utilizing representative data that accurately reflects the experiences and needs of all individuals, particularly those from historically marginalized backgrounds.
When products are not developed using diverse and representative data, they often end up being less useful for everyone. To address this issue, we have taken steps to retrain some of our earlier machine learning models using more inclusive datasets. These datasets serve as the foundation for building our hardware and software products.
**The Importance of Inclusive Datasets for Camera-Based Products**
Products that rely on cameras, such as taking a photo or using face unlock on your phone, require special attention in ensuring inclusivity. In this regard, we have made significant progress by utilizing more inclusive datasets to develop Real Tone on Google Pixel. This feature authentically and beautifully represents skin tones for all users.
**Collaboration with TONL: Capturing Skin Tones Authentically and Beautifully**
Over the past two years, our team has partnered with our colleagues from the Responsible Innovation team to collaborate with TONL, a stock photography company. TONL’s name itself pays homage to the importance of accurately and beautifully capturing all skin tones. Together, we have sourced thousands of images featuring individuals from historically marginalized backgrounds.
**Photography Across the Gender Spectrum and Diverse Skin Tones**
During our collaboration, our aim was to ensure inclusive representation across various dimensions. We sought to include photography of models across the gender spectrum, individuals with darker skin tones, and models with disabilities. By incorporating these diverse identities, we strive to address the intersectionalities of these experiences in our products.
**Expanding the Project: Collaboration with Chronicon and RAMPD**
Building upon the success of our collaboration with TONL, our project has expanded further. We are now working with Chronicon and RAMPD to source custom images that feature and center individuals with chronic conditions and disabilities. This expansion allows us to enhance inclusivity and ensure that these individuals are accurately and beautifully represented in our products.
As we continually work towards creating more inclusive products, utilizing representative data is paramount. By retraining our machine learning models with inclusive datasets, we can overcome the challenges of underrepresentation and create products that serve the needs of all users. Our collaborations with TONL, Chronicon, and RAMPD play a vital role in capturing the richness of diverse experiences, ensuring that our products reflect the reality of a broad range of individuals.