Swiggy Ends India’s Classic Food Debates with a Flawed Algorithm
In the midst of controversy, our brand did the unthinkable by Benedict Gershom, Copy Supervisor
“It is safe to assume that bias exists in all data. The question is how to identify it and remove it from the model.”
Over the years, machine learning has evolved and adapted to its environment; the more we feed it, the better it becomes, but unfortunately, sometimes it falls prey to certain societal constructs we as humans have built. Algorithmic systems are increasingly being taught about gender equality, racism, and other ethical dilemmas people face everyday, and I think we need to cut some slack when this poor machines gobbling up 1s and 0s are put on the spot.
This said, last week a controversy emerged as Twitter users noticed a biased algorithm glitch. A glitch that surfaced around the web, with several users attempting to conduct their own experiment by asking the AI a few difficult questions. The algorithm in question? Speedy Neural Networks for Smart Auto-Cropping of Images.
Swiggy took this glitch and leveraged it with a twist.
A little background:
When users upload images on Twitter and it doesn’t adhere to Twitter’s standards, its cropping algorithm scans the entire image using machine learning and artificial intelligence, identifies a certain part, and generates a preview. You too can try it: Upload a vertical format image and only a certain part appears as a preview on Twitter.
Researchers had a diabolical experiment in mind: to check if Twitter’s AI was biased and the results were 🤯. They uploaded a few images of a white and a black person and every single time, the AI cropped the black person out of the frame and previewed the white person thus sparking a worldwide debate. Twitter’s AI had a technological flaw, which has been a common occurrence amongst machine learning algorithms.
What did we do?
Swiggy, being the fun, quirky brand it’s always been wanted to see it in a different light: a different kind of experiment. Indians always have conflicted opinions when it comes to food. We pit these food-related idiosyncrasies against each other to see what the AI would choose. See for yourself:
Results:
And the AI didn’t fail to disappoint. It did stoke the fire of controversy again by choosing Pineapple on Pizza over Pizza. Interestingly, it also chose elaichi over Biryani 👀. And it turns out that Twitter AI is quite a peculiar eater who sometimes can’t make up its mind between puchka and golgappa.
Chutney Insight:
If there’s one thing digital agencies and brands can take from this post: it’s the way your brand can take advantage of flaws in technology and still have a conversation that’s true to its tone. Instead of harping on the technical inaccuracies, see how the same thing can be used by your brand to build itself on social. And if you see no brand connect, don’t do it for the sake of doing it.
