When Facebook (NASDAQ:FB) suggests you "tag" a friend in a photo, it generally suggests that friend's name. That small interaction provides a glimpse into the world of an emerging and powerful aspect of artificial intelligence (AI) in action -- image recognition. With its treasure trove of words and pictures from 1.79 billion monthly active users, it is using that data, combined with recent advancements in AI, to propel this and other technological advances.
Facebook may well have the lead in facial recognition, even extending a step further into the realm of facial verification. It released a research paper in 2014 in which it reported 97.35% accuracy, which approaches human levels of recognition. So accurate, in fact, that Facebook removed the feature for European Union citizens at the EU's request to protect privacy.
What is deep learning?
So what is the science behind these developments? Deep learning, a discipline of artificial intelligence, creates a computer model of the human brain that is combined with powerful algorithms and fed massive amounts of data. It is "taught" to recognize similarities as well as distinguish and differentiate. Simply put, it learns to recognize patterns across large swaths of data. This has led to dramatic advances in the areas of voice recognition, image recognition, natural language processing, and user profiling.
Facebook believes this science is so important, it went so far as to create a series of educational videos on the subject. These detail and define what AI is, how it works, the different methods it employs, and what it hopes to achieve. You can watch them here.
A picture is worth a lot to Facebook
Facebook is pursuing numerous initiatives rooted in deep learning to better connect with its users. Visual understanding, more advanced than image recognition, seeks to not only identify the people in a photo or video but also understand what those people are doing, as well as identify surrounding objects. Facebook plans to deploy this technology to better serve the visually impaired by providing descriptions of things they may not be able to see. Additionally, this technology makes it easier to identify offensive content, filter out click-bait, and identify something like terrorist propaganda.
Another Facebook system known as Deep Text uses linguistic understanding to comprehend content. With 125,000 comments and 400,000 news stories shared every minute, this newest deep learning system will sift through the data posted on its site every day, attempt to understand it, and use what is has learned to match users with material that will be of interest.
For instance, it ventures to understand how people use slang and determine the specific meaning of a word that may have multiple meanings by understanding the context. This technology will also reportedly help match users with advertisers, weed out prohibited content, rank search results, and identify trending topics.
AI miniaturized for your phone
Facebook recently succeeded in condensing a neural network to a size that can reside on a phone. This app will redraw your photo or video in the style of one of the great art masters, such as van Gogh or Picasso. The app is merely a novelty. What is truly groundbreaking is the miniaturization of the neural network, which typically resides on a bank of servers. This will bring real time applications of deep learning to bear in a variety of as yet unimagined ways.
The Facebook future
Facebook has approached each of these developments with an eye toward improving the user experience and finding additional ways to monetize content posted on the site. It has more than 40 teams and 25% of its engineers incorporating AI into its products and services, which it believes will enhance the user experience. It may be difficult to quantify these benefits, but continued growth in its user base continues to impress, with users up 20% year-over-year in the most recent quarter. The investments Facebook is making in this emerging technology will continue to reap benefits for years to come.