A MUCH NEEDED CHANGE IN FASHION ALGORITHMS
In the past ten years fashion has changed massively, technology has helped pushed the fashion industry to change and adapt. In 2018, Karinna Nobbs helped to create an augmented reality try-on tool for cosmetics. Whilst it was been tested a key finding was that technology like this worked perfectly fine on a younger model who was white or Asian, for black models or older models it wasn’t as successful. So to fix this, they had to retrain the algorithm so that it would recognise a wider range of faces, and then the tool would also be able to calibrate colours based on the models features. Before diversity would not have been thought of as an issue within technology and artificial intelligence or an algorithm, but this proved we need to be inclusive with tools we are building as well as everything else on the fashion industry.
Artificial intelligence surprisingly can have similar racial biases to people, after all it is created by a person. In fashion this usually manifests in a certain kind of model being selected, or in extreme cases online searches have previously mistaken a black model’s legs for a pair of jeans. If this doesn’t call for a big change, then what will? If this continues to happen, a lot of people around the world will continue to be alienated from an industry that should cater to everyone, and over time potentially reinforce racial stereotypes.
Artificial intelligence can be confusing to wrap your head around if you don’t deal with the tech side of things, what it does essentially is it allows machines to perform tasks that people would have had to sit and input into a computer-it makes things a lot quicker basically. These tasks can be as simple as reading some text, seeing images, making decisions or recommending a product. It is the algorithms that make that happen, which are basically a set of rules that are applied to some specific data. If you input a limited amount of data, that is what it will learn from and that is what it will identify; so if you input more white models into the data, it will be a lot easier to identify a white model over a black model, and then you might specifically input blue dresses into your system, it will recognise blue dresses, but what if you have a black customer who wants a red dress, what do they do? To put it simply, the algorithm would have a harder time recommending products for them.
However, biases are reinforced on a regular basis. Biases in search results are becoming more and more of a challenge because the behaviour of people using those tools online is constantly reinforcing it, so if the algorithm can’t detect something is wrong, it will repeat that action. Now there are people creating more algorithms that will give you more diverse results, but for us humans sat on the internet, we are being encouraged to review, and if possible, adjust any automated content.
This can be a hard topic to come to grips with, but what changes would you like to see?