Facebook has been under fire lately for a number of reasons. The social media giant’s four-hour shutdown was a major controversy, as many users wondered how exactly one of the biggest tech companies in the world could suffer such a disastrous hiccup.
Then, a former employee and current whistleblower revealed some of the shadier in-company doings of the company, alleging, among other things, poor oversight with regards to content regulation.
Many an opinion article was written on issues like social media’s tolerance for and even encouragement of, through inhuman recommendation algorithms, content relating to eating disorders and other bad influences on users, specifically teenaged girls.
Many people may see these controversies as indicative of the coming decline of this most titanous of Silicon Valley juggernauts, but such a conclusion is quite wrong. Though Facebook is likely ironing out the above issues with some internal changes, it has been steadily pushing forward with innovations.
One such innovation has to do with the field of artificial intelligence, which is dedicated to teaching computers to independently reason their way to goal-reaching actions.
Specifically, it is an innovation in the field of machine learning, a subfield of AI dedicated to teaching AI agents to not only act and reason on their own, but to learn on their own, enabling them to improve their performance and understand their environment without any assistance from a human.
Facebook’s new machine learning process is one that draws from innovations in computer vision, one of the burgeoning fields under the larger “umbrella” fields of AI and machine learning. Computer vision is dedicated to helping agents better perceive their environment and the processes that occur within it.
The new process goes by the name of Anticipative Video Transformer (AVT), and its basic function is to make predictions about upcoming actions in an environment.
Since it’s October, and we’re hurtling headlong towards Halloween, we’ll find our human-world analogue in scary movies. When we watch a scary movie, it can often be easy to predict what will happen next. Since characters in horror movies tend to always make the wrong next move, we can easily make predictions about whatever fate-decreeing misstep a character will make, to our cynical amusement. And why do we know this? Because we’ve become familiar with the tropes of horror movies from the other horror movies we’ve seen. We have the data, and can guess based on a perceived storytelling formula what twists and turns lie ahead of us.
When it comes to AVT, it will be able to watch a user’s real-time activity in order to make predictions about which moves would be better than others. A common use will be in augmented reality situations, where an AI agent, through computer vision, will analyze an environment and draw out the features that are worth noting, leading to more informed actions. Think of it like holding an iPad up and seeing the world like the Terminator does, scanning the environment to find the best course of action.
This will come in handy in scenarios where you will need assistance with a task you may not have much knowledge of, and Facebook offers the example of changing a car tire as one such scenario. By analyzing scores of videos and images relating to a host of processes, AVT will be able to lead people step-by-step through a variety of actions, from major to minor.
Whether you want to use AVT for changing a tire or cooking a new recipe, we hope you’ve learned something from this article. For more information about the weird and wild world of machine learning, be sure to check out any of the blogs in our machine learning series!