Facebook has never been a stranger to the public eye or the news headlines. Between being the vehicle of choice for the alleged Russian interference in the 2016 U.S. election to questioning by the U.S. senate and congress for its shoddy privacy habits, team Zuckerberg isn’t used to sitting on the sidelines. And now, the company’s latest venture is continuing the race for Artificial Intelligence (AI) mastery by using Instagram photos to train image recognition AI.
What Facebook has been up to
Facebook researchers have been running billions of Instagram photos (and 17,000 hashtags) through their image recognition AI models to teach them to recognize and make sense of images and hashtags. The goal of the “large-scale hashtag prediction model” laid out during the pre-training research was to improve Facebook’s deep learning model, which it did. The algorithm effectively parsed through thousands upon thousands of hashtags, learning to prioritize certain hashtags over others and label and categorize images on its own.
Unlike Facebook’s facial recognition algorithm that aids users in tagging photos, this new round of AI experimentation is focused on objects. Despite the extraordinary amount of data that was analyzed (the largest data set ever analyzed by AI), the algorithm was able to identify objects in images at 85.4 percent accuracy, the highest yet.
What exactly is deep learning?
Deep learning utilizes algorithms that are modeled after the human brain and neural networks. Deep learning and machine learning are sometimes used interchangeably, with the goal being to mimic human behavior—that is, to learn from experience through supervised and unsupervised learning. But in deep learning, a machine’s algorithm essentially learns to make decisions on its own, without the guidance of a programmer.
Is this a privacy breach?
Back in 2016, about 87 million Facebook users reportedly had their data shared (without their consent) with political firm Cambridge Analytica. And Facebook’s latest data dance doesn’t come without questions. While Instagram photos that aren’t set to private are essentially public data, (meaning no legal boundaries are being pushed), is it right for Facebook to analyze and use photos in their research without permission? Time and public opinion will tell.
Facebook’s former failed AI experiment
This certainly isn’t the social media giant’s first attempt at AI. Last year, the company was training AI chatbots to learn the rules of a game and how to negotiate. But when “Bob” and “Alice” went off-script and started speaking what amounted to gibberish by human standards, Facebook reportedly shut the two bots down.
Key takeaways and implications
While nobody can yet say what long-term implications the research and models could have down the road, one thing is certain. The research has laid a vital foundation for researchers and futurists to develop AI that could sort through and make sense of masses of “messy real world data.” The researchers led the largest test of its kind, with the greatest degree of accuracy thus far. And beyond recognizing #dogs or #cats, the applications of this test could extend to aiding in the development and fine-tuning of autonomous vehicles, which depend on image and object recognition.
In an interconnected world, some fear it’s just a matter of time before intelligent machines realize how everything is connected and use it against us. Will Facebook’s continued race to stay ahead of the AI curve usher in the dreaded future which tech giants like Elon Musk, Bill Gates and Steve Wosniak fear? AI certainly can change the world. But the question remains, will it be for better or for worse?