YouTube has become a platform where anybody with a camera and a personality can amass a following and turn content creation into a job.
This is made possible through advertising or rather AdSense. YouTube runs advertising on videos and if you are a partner, a percentage of the advertising revenue gets shared with you. The more views you get the more money you make. It’s a simple equation but lately it has been a lot less simple.
Demonetisation has been a common topic among YouTubers for sometime now with the core of the conversation being around what causes demonetisation.
One channel that has made it its job to answer that question is Nerd City.
While it might be a small channel with 690 000 subscribers, the work these folks do is incredible and this morning it published a very interesting video.
Bearing the title “YouTube’s Biggest Lie” the team at Nerd City together with YouTube Analyzed and Sealow have discovered that despite YouTube encouraging LGBTQI individuals to create content, that content is often instantly demonetised.
The video reveals that video titles containing words like “gay” or “lesbian” are instantly demonetised, while videos with “friend” replacing “gay” or “daughter” replacing “lesbian”, weren’t.
How was this tested? Nerd City explains that it uploaded several short videos to test its theory. The videos contained no content that could flag it for demonetisation so only the titles were being tested. As many as 15 296 words were tested to see how the platform would rank them.
The theory (keep in mind that YouTube has never revealed how the website works) is that words are scored according to their severity. This score ultimately affects whether a video is demonetised or not.
Now, at this point folks would likely point to YouTube’s algorithm, artificial intelligence and machine learning as the root cause of this but Nerd City discovered something rather interesting – bias.
We’ve touched on this topic before but to refresh – training AI requires an unbiased team. Naturally, nobody is unbiased so experts often point to diversity as being as important.
While training an AI, the person behind the screen could for example, weight “gay” more heavily than “straight” resulting in the problem YouTube faces now.
One of the firms YouTube uses to train its AI, Lionbridge AI operates in several countries where homosexuality is illegal or frowned upon.
This is not to say that YouTube only uses one firm for its AI or indeed that the AI used by YouTube contains those biases, but it does raise an eyebrow and warrant a deeper look.
The fact of the matter is that by testing 15 296, Nerd City and friends have found something deeply concerning that flies in the face of YouTube’s push to get more LGBTQI creators.
You can find a list of words that were tested over at this link with green words meaning they can be used without threat of demonetisation and yellow meaning you will be demonetised.
You can watch the full video from Nerd City below just know that it clocks in at 30 minutes, so perhaps save it for your lunch break this afternoon.