In the realm of technology, embracing diversity is more than just an ethical mandate. It’s a strategy that shapes the usefulness of products, ensuring that they cater to a wide array of individuals. In a remarkable endeavor to foster inclusivity, Google has started employing more diverse and representative data to refine its machine learning models. The objective? To build products that effectively resonate with individuals hailing from historically marginalized backgrounds.
Google’s pursuit of inclusivity is making waves, particularly in the creation of software and hardware products. These products often rely on machine learning models for their performance, and in turn, these models heavily depend on datasets. When the datasets are inclusive and representative, they help build products that work well for all users. This impact is especially potent in camera-reliant products, such as face unlock features and photo-capturing applications on phones.
Artificial intelligence training data
Take, for example, the Real Tone technology on Google Pixel. It’s an innovative result of Google’s commitment to inclusivity, ensuring all skin tones are portrayed authentically and attractively. This technology was brought to life using diverse datasets, enhancing its ability to serve all users efficiently.
But Google didn’t embark on this mission alone. Over the past two years, Google has teamed up with TONL, a renowned stock photography company, and its Responsible Innovation team. These alliances have given Google access to thousands of images featuring people from diverse backgrounds, encapsulating a spectrum of skin tones, genders, and disabilities.
Furthermore, Google’s partnerships have extended to Chronicon and RAMPD, helping the tech giant source custom images that portray people with chronic conditions and disabilities. This expansive and inclusive image collection aids Google’s product teams in identifying potential fairness issues while shaping their machine learning models.
AI representative training data
At the heart of Google’s inclusivity drive is a unique dataset: The Monk Skin Tone Examples (MST-E). Curated by the Google Skin Tone Team in Responsible AI and TONL, this dataset offers images and videos of 19 individuals, with skin tones varying across the 10-point Monk Skin Tone scale.
The MST-E dataset is designed to be a practical tool. It helps teach human annotators to test for consistent skin tone annotations under diverse conditions, which is key for the performance of AI-driven products. With its help, Google can ensure its products are attuned to the needs of individuals of all skin tones.
Looking ahead, Google is committed to continually refining the representation in its datasets. The company’s ultimate ambition is to build the most inclusive and equitable technologies for every single user. This venture isn’t just about creating tech products—it’s about sculpting a more inclusive future where technology serves everyone, regardless of their background.
So, if you’ve been wondering how tech companies can foster greater inclusivity, Google’s efforts offer a compelling blueprint. By leveraging diverse datasets, the company is not just refining its technology, it’s also reshaping the way we understand and experience inclusivity in the digital realm.
Hopefully we can look forward to a future where technology not only acknowledges our diversity but also celebrates it, crafting a more inclusive and representative digital landscape for all to use and enjoy.
Source & Image : Google
Filed Under: Guides
Disclosure: Some of our articles include affiliate links. If you buy something through one of these links, TechMehow may earn an affiliate commission. Learn about our Disclosure Policy.