The new AI project, which Facebook calls SEER, was able to see and recognize over one billion public images on Instagram. While SEER is currently only a research project, there are plenty of applicable uses for this kind of AI on social media, from accessibility to content moderation. “Facebook could use this model to build user-facing products powered by AI for use cases,” wrote Matt Moore, the vice president of product management at Zype, in an email to Lifewire.
The SEER Technology
Facebook said SEER (which stems from SElf-supERvised) was able to outperform existing AI models in an object-recognition test. According to the social media company, SEER was able to achieve 84.2% accuracy in image tests. Facebook said it’s focused on a kind of AI tech that can learn independently, without the help of an algorithm. “The future of AI is in creating systems that can learn directly from whatever information they’re given—whether it’s text, images, or another type of data—without relying on carefully curated and labeled data sets to teach them how to recognize objects in a photo, interpret a block of text, or perform any of the countless other tasks that we ask it to,” Facebook’s researchers wrote in the blog post. Moore further broke down how SEER differs from AI tech we are typically used to. “The biggest differentiator of this new SEER model is that Facebook is using a very large volume of raw data and letting the model train itself—as opposed to manually curating models with limited datasets,” Moore said. He added that using a raw dataset could give more accurate recognition predictions in the real-world. “Raw datasets can also help reduce biases built into recognition models built off of limited datasets,” Moore added.
How SEER Can Be Used
For now, SEER is only a research project. Still, experts say SEER’s development could pave the way for more versatile, accurate, and adaptable computer-vision models, while bringing better search and accessibility tools to social media users. One tool, in particular, that could greatly benefit from this technology is generated text for describing images to people with visual impairments. “Alt-text is a field in an image’s metadata that explains its contents: ‘A body standing in a field with an elephant,’ or ‘a dog on a boat,’” wrote Will Cannon, CEO of Signaturely, in an email to Lifewire. “The enhanced system should be a feast for visually damaged users, and may assist you in finding your pictures quicker in the future.” Other useful applications for this technology could include better automatic categorization of items sold on Facebook Marketplace and more accurate systems to identify harmful images. “Facebook’s AI could automatically identify and remove sensitive video content that violates platform terms of service, creating a healthier community,” Moore added. Even though Facebook previously has been in the hot seat for using its facial recognition technology without the consent of some of its users (specifically in its photo-tagging feature), experts say this AI poses little to no threat to your privacy. “With the power this model holds, Facebook has open-sourced a library for the public to inspect, but the Instagram users’ image data used to nurture the AI will not be disclosed to the public,” wrote David Clark, attorney at The Clark Law Office, in an email to Lifewire. “This preserves the use of user data to benefit only projects fully sanctioned by the company.” Clark added that, ultimately, when you sign up for Facebook and Instagram, you allow the images you upload to be in the company’s authority. In the grand scheme of things, using your photos and data to build better software is one of the better things Facebook can do with your data, he said. “This project only means that a sea of image databases is opened up to the larger computer-vision community to advance developments that will give birth to better software and programs,” Clark said.
