How does Facebook use AI? The core of Facebook’s business is learning about us, its users and collecting that information for advertisers. In return, we get new features and improved functionality that help the social network to learn more about us, down to the tiniest detail. It’s an ongoing sequence, much like the natural water cycle, where every party is satisfied with the outcome. However, Facebook AI is close to having a massive two billion active users and that means mountains of unstructured data. The question is – how do you make sense of these enormous amounts of data? By using artificial intelligence, of course.
While there is no obvious way to communicate with the AI part of Facebook like voice commands, that doesn’t mean it isn’t there. Much of the Facebook AI work is done behind the curtains, be it through the News Feed, images, video or chatbots. As the number of users rises, so will the level of automation. Artificial intelligence helps to structure the huge sets of data and turn them into actionable insights. Essentially, AI is used as an analytical tool to help Facebook provide better service and learn more from its users, while naturally getting something in return.
Today, a great deal of artificial intelligence powers the way the social network engages us. Much of it relies on a technique called deep learning (DL). It uses neural networks, essentially software systems that copy the way the human brain works and learns. Deep learning enables machines to break down data into basic elements and learn to classify it on their own. The reason why DL is so useful to Facebook is that it can be exponentially applied to more and more data that we share as the algorithms become more sophisticated. The fact that Facebook studies the activity of its user base is no surprise. What might be is the way they do it.
Facebook AI for text analysis
It all starts with text input. While the social media platform has progressed far beyond mere typing, the text is still a rich source of information. Large sets of data on Facebook are still text-based through comments, posts, and stories. For that reason, the company developed a tool called DeepText that analyzes text on the platform contextually. Through numerous neural networks, DeepText achieves near-human accuracy by correlating words in a specific content. The tool itself doesn’t have a fixed meaning of every word, which allows it to understand its meaning and change it depending on other words around it. That way, it learns on its own how we use certain words by assigning labels to them. As a result, the technology can now understand over 20 languages and apply its learnings from one language to another.
As for the practical use, the tool is utilized to steer users towards products and services they might want to use, based on their conversations. It draws on its advanced algorithms to decide what is the appropriate response, depending on the context. In addition, for the past couple of months, Facebook has been testing recommended responses in Messenger. These AI-powered suggestions offer different actions based on your chat. For instance, sending your current location if someone asks for it or offering a selection of stickers to send as a response to a message.
Facebook AI for facial identification
For basically knowing what each user looks like, Facebook uses another DL tool called DeepFace. In business since 2014, DeepFace has a scary-good accuracy of 97 percent success. To put that in perspective, comparative testing showed humans had lower accuracy by one percent, meaning this is one layer of artificial intelligence that is actually better than us right now. Obviously, such a sophisticated image recognition tool is ideal for various uses related to security. Facebook also believes the tech could greatly benefit the visually impaired. By annotating images with text, they become more accessible to visually impaired users with richer descriptions, as well as easily searchable by all.
However, image recognition does carry a bit of controversy as it treads on privacy freedoms. The social media giant has its fair share of dealing with privacy cases and regulations, which is perhaps why there is so little news regarding DeepFace since its launch. In the meantime, Facebook has open-sourced the software in hopes to accelerate its application to video. The company believes that making the technology publicly available is crucial to its further advancement, including live video and real-time classification.
Facebook AI for content targeting
This is happening for quite some time, even if it goes unnoticed for most of it. Much of the content that appears in our News Feed is thanks to AI algorithms. These state-of-the-art algorithms, as Facebook likes to call them, deliver the most relevant content. They rank and personalize stories, filter out anything offensive, point out trending topics and much more. A fairly present example of Facebook AI engagement is learning it to identify fake news. Facebook experimented with training machine learning algorithms through human input. A user would identify common phrases found in various clickbait stories. The system would then recognize and downgrade future stories that featured those phrases.
Ultimately, what you see browsing Facebook might depend on what AI thinks you would like, not your circle of friends. As a rule of thumb, the social network displays content your friends have interacted with in one way or another, be it by sharing or liking or commenting. Nevertheless, Facebook also uses engagement by your friends as a guideline on what might interest you. If Facebook AI understood the content, it could make a rational decision whether something might be of interest to a user who previously engaged with similar content and show it, despite a friend sharing it. Naturally, all of this hinges on AI’s thorough understanding of the text, photos, and videos.
AI for ad targeting
The same principle applies in this case. This is perhaps a more obvious way of utilizing artificial intelligence that benefits both Facebook and advertisers. The social media giant is using AI to filter out which ads to display to which users. AI determining which content will appeal the most to which users leads to better-targeted ads. It’s not surprising to see Facebook delving into AI-based solutions as advertising is essentially the foundation of its whole business. With big user numbers comes a big potential for gaining a competitive edge in a market filled with high-profile competitors like Google and Microsoft. Having artificial intelligence learn as much as it can about us and serve it as an actionable insight makes reaching the right audiences much easier.
Conclusion
Facebook provides a service to nearly a quarter of the world’s population. That makes for a gigantic user database which simply needs artificial intelligence to power and maintain it. Thanks to Facebook AI involvement, the social network’s capabilities in terms of delivering the most relevant content have improved in recent years. We now have a sophisticated image recognition system that can automatically identify actions and find objects in photos even if they don’t have a tag or a caption already. The social media giant is actively developing Facebook M, its vision of a digital personal assistant. It’s focusing on and investing in AI development more than ever. To sum it up – Facebook is all in on artificial intelligence.
The company has a good starting point. It’s dominating the social market with its own platform and Instagram and looks to transfer that dominance onto artificial intelligence market. It’s a race with other tech companies to best utilize what is fundamentally the future of computing. We are still in the early stages of AI application so every breakthrough is a step closer to the end goal. In that regard, deep learning will continue to be a key part of developing comprehensive Facebook AI. However, it’s not just Facebook that will benefit from it in the long run – numerous others will also take advantage of it, either through Facebook’s platform or its open-source support. It’s a long and exciting journey we are about to witness.