Twitter Rolls Out Automated Porn Detection
LOS ANGELES — The back-and-forth frenzy over a possible porn crackdown by Twitter has a new twist, with the revelation of its new artificial intelligence engine, dubbed “Cortex.”
Considered by adult marketers and performers to be the most “porn friendly” of the mainstream social media services, Twitter has connected millions of fans with their favorite adult entertainers, who rely on this exposure to cultivate audiences and to attract new roles.
It is no surprise then that the industry trembles with each rumor of an impending crackdown on adult content by Twitter.
Now, the popular social network has a new weapon in its arsenal that will make it much easier and more cost effective to identify material that it deems “not safe for work.”
What it does with that data remains to be seen, however.
Twitter’s acquisition of MadBits, an AI firm specializing in identifying the objects seen in photos and videos, led to a mandate to develop a system for tagging NSFW images.
“When you do an acquisition — even though they’re coming in to do something broad — you want to give them something specific, so you get to know each other and make sure the acquisition works,” Twitter’s engineering chief Alex Roetter explains. “So we gave them the problem of NSFW.”
That system is now reportedly in place, scrubbing images in real time and replacing legions of manual image reviewers (a figure of 100,000 workers has been claimed) that tediously scanned the content posted by Twitter users for offensive materials, that include not only sexually explicit materials, but violence and “hate speech.”
“According to [MadBits co-founder Clément Farabet], if you tune the system to identify about 99 percent of all porn and other objectionable images — allowing the company to warn users with interstitials in the Twitter timeline — it will incorrectly flag perfectly acceptable pics just seven percent of the time,” Cade Metz wrote for Wired.com. “These numbers are entirely dependent on Twitter’s definition of NSFW, of course.”
In addition to scrubbing feeds for NSFW content, Twitter Cortex drives machine learning across the company’s network to improve the user experience, enabling greater relevancy in advertising and recommendations for new contacts and more targeted tweets — something that Twitter already does — but which will become much more accurate with Cortex.
Twitter Cortex is a neural network that “learns” through experience, for example, by analyzing images that are tagged by humans in order to identify what is or isn’t NSFW — improving its accuracy over time, until the human element becomes redundant.
“You need humans, generally, to label the data,” Roetter notes. “But then, going forward, the model is applied to cases you’ve never seen before, so you dramatically cut down the need for people. And it’s lower latency, of course, because the model can do it in real-time.”
Metz notes that other sites and social services already employ neural networks on a massive scale, with Facebook using the technology to identify faces in uploaded photos; Google using them to aid in voice recognition; and Microsoft translating live Skype conversations from one language to another.
While it remains to be seen whether Twitter will clamp down on adult oriented materials posted by its users or not, what is clear is that today, it would only take the flip of a switch…