I’m learning that sometimes with blogging it’s not about getting the perfect thing out. It’s about getting something out when you told yourself you would. Today marks exactly 4 months until my 25th birthday (well, Feb. 21st did) and just over 8 months in the Bay. I’m reflecting and vision setting. Moving here is the second decision I ever really made for myself, and just like the first, every part of it has been amazing.
Even when it’s sucked it’s been great, because it’s been mine.
I am proud of myself for believing in me, and giving myself the space and time I need to grow. I am making choices for my life that feel right for me, and I am so grateful for everything I have attained, and everything that’s coming next.
Anyway, here’s a piece I wrote for work. I hope you enjoy it and also that it teaches you something. If you have questions please ask. Understanding this work and how to communicate that understanding to other people is part of the vision setting.
Thanks y’all, always.
Response to Facebook is using billions of Instagram images to train artificial intelligence algorithms (The Verge – May 2, 2018)
Lately, all my homies on Instagram are into a new filter, which I’ve colloquially named ‘The Headband.’ The Headband works by randomizing a series of images in response to a question, and then projecting the final result above your forehead. While watching a friend’s IG story, I had a startling thought.
What if The Headband is being used to improve Facebook’s facial recognition algorithm?
It’s not a thought I would have had three months ago, before joining AI4ALL and being thrust into the vaguely dystopian present that is the AI industry. However, as I watched my friend, staring directly into her camera’s lens for a full five seconds as The Headband randomized the results of her query, I couldn’t argue that if you wanted to get clear pictures of someone to improve an algorithm’s accuracy, there was hardly a better method, short of explicitly asking users to upload selfies.
Would Facebook store data like this without our permission, though? Possibly. And in Facebook’s case specifically, there’s precedent.
The Verge reported in 2018 that Facebook had begun using Instagram hashtags to label images, removing the need for human labeling, and allowing the billions of user-upload images in its database to help train its computer vision and object recognition models. At the time, algorithms were trained primarily on object-based data and used to help moderate content on Facebook’s various platforms (Instagram, WhatsApp, Messenger); however, Facebook is one of the first public adopters of facial recognition, which has powered the platform’s suggested tagging feature for years.
Ok so Facebook is, in all likelihood, using our images to train its algorithms. Why does it matter?
While there are significant privacy concerns in co-opting user data to build profitable algorithms without their consent, and we all know how important image labeling is in mitigating algorithmic bias, what concerns me more is that the average user has no basis to understand what a computer vision algorithm even is, let alone how it’s trained or how its encoded biases are impacting their lives. The larger issue with corporations assuming ownership of user data is that most users don’t even know that their data is valuable.
If I had come across this article two years ago when it was written, I might have shrugged, if I read it at all. Terms like ‘image labeling’ and ‘data mining’ only mean something to the subset of the population either designing algorithms or aware that they exist. And unfortunately, this intersection doesn’t often include the people most impacted by these algorithms’ development. If we are going to create a future in which the most marginalized groups in our society are uplifted, rather than further disenfranchised, by AI technology, we have to start by teaching them that the technology exists.