Diversitydata.IO

Design Process

In looking to find an API in support of Data Feminism principals, I came across an API called diversitydata.io that sounded quite bold and problematic in their claims. Diversitydata.io claims to be able to determine gender and ethnicity simply based on the first and last name. It seems very heavy handed in touting their AI generated results and their use cases, which was suppose to encourage diversity.
I doubt this API is taken seriously by anyone but I wanted to dig a bit further in API’s that claim to do similiar work. I found one called NamSor API which seems more accurate, since they look into the origins of the name. I reached out to NamSor about using their API and explained why I wanted to use their API. The person I spoke to forwarded me a google sheet with different names, ethnicity and gender that used Diversitydata.io’s API. I find it strange that Diversity IO’s API was referenced by anyone but themselves.
For my project, I wanted to focus on the problems of their claims. My first thought was to represent the flaws and reflect on how this fails to follow principles of feminist data. That seemed too obvious and uninteresting. I decided the best approach was to double down on their claim. I found a site that generates AI faces that had matching categories generated by diversitydata. With this latest implimentation of their API, not only will you get the gender, and ethnicity, you will also get a picture of the person. Link to diversity API Link to AI generated face

Initial sketches

Reflection

An appropriate way to approach this project to loop back to the begining of this class. One of the first questions we were asked in this class is if technology is neutral. Though this touches a few other principles, I wanted to answer this based on the 6th principal of feminist data. Consider Context: Data feminism asserts that data are not neutral or objective. They are the products of unequal social relations, and this context is essential for conducting accurate, ethical analysis. In plugging in my name, it says I am a white female. I tried a few other names and it did not get much more accurate. I suppose it is a good thing since if they were accurate this api would be used to do the opposite of what they claim it does. The only way this API could be accurate was if you put in traditional names (these arent always accurate either). Though I still think no one will take this seriously, the issues it creates seem appropriate to be explored. One of which is the tendency to trust and overstate the power of technology. As we are aware, currently machines are far from perfect and algorithms are based on limited resources. This can lead to real problems like the biased recidivism algorithm. In a way the algorithm of diversitydata carries very similiar traits to the recidivism algorithm. On the surface they both carry an overstated capability based on non biased ideologies. The core of their product will however be for division. The only factor that makes diversitydata.io lean on the side of irresponsible instead of malicious is their overt transparency.