I’m not sure you understand what they’re trying to do. The purpose is not to accurately guess a person’s race from their first name. They know that some names practically specify race and the purpose is to determine if there is a correlation between those names that imply a specific race and the number of declines for guests with those names. They also want to find the correlation between profile photos that imply a specific race and number of of cancellations for those guests.
While I have no idea about the names of population as a whole, my example simply shows that half of my guests that were Black have names that imply their race, so half of them might fall victim to discrimination under Airbnb’s current policy of showing only the first name before booking. It should be obvious that that would also apply to some percentage of guests that are Asian, Hispanic, and Middle-eastern, etc.
“Guessing” is exactly what they want. Based on what I read, this is what I think they are doing. For every guest that opts in, they are going to take the first name and show only the name to a bunch of humans. Then they’re going to assign a score to that name that indicates the perceived race based on what race the humans guessed. After the score is assigned, every time a guest with that name requests to book and gets accepted or declined, it will be logged for statistical purposes so that it can be compared with the acceptance rate of other names.
They will repeat this for the profile photo, assign a completely independent score, and log that with host cancellations.
What’s important to understand is that it’s completely irrelevant if the people that originally guessed were correct or not, as long as the number of “guessers” is representative of the population (specifically, it should be representative of the population of Airbnb hosts in the US). All that matters is the perception offered from the name (or photo).