Artificial intelligence is embedded deep in our lives. Much of the focus of the world’s share of tech ingenuity is aimed toward developing AI-powered products and services. Solutions that are more precise, intuitive and clever on their own are at the forefront of many projects around the globe. Not one to be outdone, Facebook is working on artificial intelligence. The social network giant has its own Facebook AI research program called Facebook Artificial Intelligence Research (FAIR). The general principle behind FAIR is pushing advancement of machine intelligence in order to create new AI-powered solutions for communication. Or, as they like to call it, “to solve AI”. FAIR is one of several AI research areas that excite the curiosity of Facebook. Applied machine learning, connectivity, natural language processing and speech, human-computer interaction and user experience are some of the areas Facebook has a strong interest in.
If you need further proof of how much artificial intelligence is prevalent in tech environments, just look at the one Facebook user we all know. Mark Zuckerberg, the face behind the most popular online social media, recently built an AI agent to run his home. It uses a number of artificial intelligence techniques, including the ones utilized the most – natural language processing, speech recognition and face recognition, as well as reinforcement learning. According to Zuckerberg, he has spent 100 hours building Jarvis (apparently, the uniform name for an AI platform) last year. The end result, per Facebook CEO’s claims, is a system that understands him pretty well and performs a wide array of functions. However, Jarvis is far from a truly advanced system which can learn skills on his own.
Zuckerberg’s Jarvis and the role of Facebook AI
The point being is that without a fundamental breakthrough, artificial intelligence is still far away from the vision of an all-encompassing entity. The things we have today are a good indication of what AI might do for us in the future. Nevertheless, that is only a small part of its overall potential which still doesn’t get us near understanding the central issue – figuring out real intelligence.
While building Jarvis was primarily a fun challenge for Zuckerberg, Facebook is actively pursuing AI development with a well-defined strategy. For the company, artificial intelligence has become more of a guideline, a foundation on which to build and create, rather than a means to an end. The social network firm is engaged in a variety of infrastructure projects aimed at better connectivity. The nature of these projects varies from company-oriented goals like maximizing the number of people with internet access (and therefore enlarging its user base) to consumer-ready products like visual interaction through stylized filters and virtual or augmented reality. However, perhaps the largest impact on the vast majority of Facebook users has the company’s platform itself. The core of its business and the starting point for all of its users, Facebook is using its chat platform to advance own AI plans to a large crowd.
Facebook treading into AI assistant territory
This summer will mark the second birthday of Facebook M, social network’s counterpart to Google Assistant, Alexa, Siri and all other personal assistants. So far, the assistant is in demo mode and only available to a handful of testers. Facebook intends to use artificial intelligence to teach it how to interpret comments, stories, images and videos. The social network stores that data for improving the relevance of News Feed content and ad targeting. This is where M enters the picture as all that data is also the basis for a cutting-edge Facebook AI assistant. While it is currently text-based only, M is learning with human input/trainers to get the gist of the user’s intent.
Here’s how the whole process goes. M performs calculations on each given question and ranks the level of probability in terms of accuracy. This means that if it estimates a high probability of accuracy, it provides the response directly to the user. On the other hand, it requires human assistance on the low probability of a not quite accurate response. When the trainer intervenes to help M understand the intent, the digital assistant records the input and learns. That way, it improves its accuracy when understanding the purpose of the question the next time.
Facebook M
FAIR has a big role in the way Facebook M interacts with users. It created Memory Networks, a neural net that collects information and transforms it into an action. In essence, it can understand and recall a specific set of data and answer questions about it. This relatively sophisticated question answering allows the AI to get better over time. It’s a step closer to understanding the relation between objects and topics, one of the main obstacles to teaching machines how to reason like humans.
In terms of Facebook M, understanding how people, objects and time relate to means ultimately reaching human-level of intelligence. Once ready for public launch, M will be free to use and a vital part of Messenger. It will improve News Feed, help out with different tasks and find information with better search results. As technology advances, M will eventually scale up to more complex reasoning. Identifying subtle intent and detecting nuance from your comments, recognizing faces of you and your friends that appear in videos (we already have a pretty good photo recognition) and interpreting your feelings and expressions and portraying them as avatars in VR environments – these are all potential applications of the end-goal vision of M.
What Facebook is effectively working on is the generalization of AI, building generalized systems that transfer knowledge from one task to another. Figuring out how to distinguish content will provide a faster response in delivering the right information. It could digest the sharing patterns that characterize fake news and use its machine learning practices to weed them out. Potentially, it could also empower real-time analysis of what you write, say and see. For now, the aim is to have M learn from trainers and become clever enough for new undertakings. Facebook is slowly increasing the number of trainers until it deems M ready for a fully automated service.
New Facebook M Features
In June, 2017, Facebook rolled out new features to help users be better friends. Now, Facebook M will remind you if it’s the friend your chatting with’s birthday. It will also remind you to save videos, articles and other pieces of content your friends send you in Messenger. M will also recognize if you’re talking to someone about making a phone call, and which case the app will pop up a “call” button for you to call immediately.
Conclusion
Despite all the breakthroughs and advances, AI is still in its infancy. With M, Facebook is trying to create an intelligent conversational agent capable of almost anything. What many often forget is that artificial intelligence is not solely about interacting with a conscious machine. It’s also about going through piles of data to make better products and services. If anyone has data about us, it’s Facebook.
Facebook has the means of delivering AI to the masses. M is just one of its many projects with a purpose to make artificial intelligence the next critical Facebook platform. However, it’s still in beta testing stage and being trained and supervised by real people. At the moment, our questions often pose too much ambiguity for AI to understand. M’s automatic text understanding and reasoning in many cases are closer to guessing than a high probability accurate answer. Thus, before artificial intelligence completely replaces the functionality of human trainers, there is plenty of work to do. So far, there is no date on when that might happen. Meanwhile, we are eagerly awaiting for a public rollout of Facebook’s AI-powered personal assistant, to see if it comes close to its promise.