Speaker 1: The world's population is much more than just fair skinned
Speaker 2: For many people of color. It's not very common to see themselves reflected in physical and virtual worlds. Sometimes a Google search will only show images of people with lighter skin, for example, or a filter will wash out someone's features. That's why Google says it's now using the monk skin tone scale, a 10 shade scale. That'll be built into its various products over the coming months. The goal is to help Google better judge and present a range of [00:00:30] skin tones. So what is the monk skin tone scale and how could it make the digital world more inclusive? We went to the source, the creator, Harvard professor, and sociologist Ellis monk, who spent over a decade studying how skin tone and colorism affect people's lives. He developed this scale based on his years of research, which Google will now be incorporating in its products and services.
Speaker 1: The main aim there is to have, have a scale that will be able to reliably [00:01:00] classify the different skin tones of people in society.
Speaker 2: A few years ago, a member of Google's responsible AI team reached out to monk after learning about his research on colorism. They looped him into the company's conversations about skin tone, equity and tech, and learned about the scale he developed.
Speaker 1: And it's certainly something that I've dealt with, you know, in my own life where you, you walk up to a hand sanitizer and you put your hand under it and it doesn't work, or, you know, there's facial recognition technology and the camera on the phone [00:01:30] and it doesn't quite work, right. Or you take a picture and the skin tones of your family members, some people look like they're well lit. Other people you can't see as well. These are all indications of technologies that were not built and tested with intention that they would work well across the entire skin tone continuum. So I just think this is a great opportunity and I'm very happy that Google decided to move forward with this project to fix these issues.
Speaker 2: Here's an example of how you'll see the monk scale show up in Google. If you're searching [00:02:00] Google images for makeup, you'll see an option to refine the results by skin tone. So you can find what works best for you. Google says it'll be implementing the new color scale in a range of its products and services. Google photos will also use what are called real tone filters, which are designed to work well across skin tones and were evaluated using the monk scale until now the industry standard for categorizing skin tone has been the Fitzpatrick scale. A six step scale created in 1975 [00:02:30] by Harvard dermatologist Thomas Fitzpatrick.
Speaker 1: And the scale has really developed around thinking through, well, if we're using UV, uh, raise as a way of treating things like psoriasis and eczema, how much in terms of a dosage should we use? You know, how do people tan? How did they react? So it really, really wasn't used, uh, the use case for that scale was not really about the classification of different skin tone.
Speaker 2: The monk scale would do a better job of representing the vast [00:03:00] range of skin tones that exist. It has just enough points to be inclusive without being overwhelming.
Speaker 3: It's not reflective of the full diversity. You know, for example, makeup scales often have dozens of different colors, but it's a small enough number that you can use it in practice in the real world. If you're a researcher or if you are perhaps, uh, somebody at a tech company, who's trying to train an AI system, an artificial intelligence system. So it's got more representation in the previous system, not the full range, but [00:03:30] it's still much more useful.
Speaker 2: Monk also noted a 10 point scale is the sweet spot.
Speaker 1: The main idea behind the skin tone scale was to have something that is optimal in terms of how large the scale is. You don't want too many points where people who are annotators can't really use it cuz you can't really differentiate, uh, because there's too many, uh, points along the scale. Uh, but also something that in terms of this color selection is really good at differentiating the dynamic range of skin tones [00:04:00] that you'll see particularly among people with color,
Speaker 2: The scale is designed so that there is never a need to expand it
Speaker 1: With the scale which you want to do really is you wanna optimize the set of points on the scale. You want to make sure that your color selection is solid. You want to test it and validate it. And we're doing more to validate outside of the us. So we've done a lot of that validation research in the us. Uh, we're gonna be doing that in India, Brazil, other countries around the world, other regions. Uh, but once you have something that seems like it's, [00:04:30] you know, representative and it works, a lot of the effort is really about implementing it. It's not so much about tweaking and changing it because it's a standard
Speaker 2: Silicon valley has often been criticized for primarily catering to the needs of white men and overlooking women and people of color.
Speaker 1: The reality is most of the people in Silicon valley doing this tech work, it's not the most diverse representative of the United States or even the world population of people. So it's very easy to kind of just have your blinders on and make a [00:05:00] product and test it on yourself.
Speaker 2: When it comes to artificial intelligence in particular, there is concern. The technology can reflect biases in the data, used to train it in one of the biggest cautionary tales Google's photos app in 2015, labeled a picture of two black people as gorillas. The label showed up in a feature that automatically categorizes photos like cars or beaches. So they're easier to search Google and other tech companies are working to prevent more of those embarrassing mistakes and to ensure their services [00:05:30] cater to everyone as part of its effort to make search more inclusive, Google will be developing a way for creators brands and publishers to label their web content with attributes like skin tone and hair, color, and texture. This will make it easier for search engines to understand what's in those images and surface them whenever relevant, Google will also openly release the monk scale. So anyone can use it for research and product development beyond Google.
Speaker 1: At this point, the tech industry, you know, the Facebooks of the world, [00:06:00] Microsoft, you know, apple, they can take the scale, run with it, use it in their products and they're ready. They're good to go.
Speaker 2: It could take a while for other companies to switch over to more inclusive skin tone scales, but heightened awareness for diversity. Following movements like black lives matter means this is an issue that's become more top of mind.
Speaker 3: I don't think it's an easy switch to embrace this in the AI world at large, but I think there's a need for it. They've done some credible research for it. There's a lot of impetus to move in this direction. So [00:06:30] I don't think it will be a fast adoption, but I think they definitely could get some other people on board.
Speaker 2: That means this could be just the beginning of a wider shift to more inclusive technology.
Speaker 1: It's not every day that you're able to, you know, make a change in technology, uh, and affect the products that really billions of people use every single day.
