This is a common opinion (which I don't agree with) that is worth addressing directly, and it has consequences for how we understand and use the sorting capabil

This is a common opinion (which I don't agree with) that is worth addressing directly, and it has consequences for how we understand and use the sorting capabilities of AI. The binary outcome consequences of misidentifying a dog breed in an app are not really important, but the ability for the breed to be a factor in informing some other aspect of a model is useful to businesses that optimize cash flows. For example, I love stafforshires, and I also recognize that a machine learning scheme that added them as a factor in a credit risk evaluation might disadvantage a number of people, but be more profitable than a model that did not include it. If this is offensive, consider that breed specific legislation already allows you to be denied home insurance (and consequently, mortgages and equity loans) based on having one of these dogs today. The code in this project could be used to mine social media images for breeds correlated with credit risk factors, and offer interest rates for the owners accordingly, and a credit instrument portfolio that incorporated this factor could outperform one that did not. In regions with breed specific legislation, someone with an banned dog has already demonstrated a deviation in their risk behaviour. Shelters are not full of failed show dogs. In fact the majority of them are mixes, particularly mixes preferred by people who were guided by sentiment over more sound reasoning factors when they got the dog in the first place - hence why the dog is in the shelter, and why they probably can't get credit on favourable terms. Is this just or ethical? No, it's predatory, but that's the social media bargain, and projects like this are worth keeping track of if you like dogs, and access to credit.