Facebook is beginning to take a new approach to the task of digitizing our personalities and spinning them into nicely saleable little packages. After all, posting and updating an exhaustive list of your favorite bands tends to lose its appeal around the end of adolescence; if Facebook wants to truly understand our material wants and needs, it will have to look into our much more personal, much less intelligible communications. This week, Facebook’s chief technology officer said the company’s newly formed AI team has its sights set on building neural networks to learn about your personality in a new and remarkably human way.
A simple model for a neural network.
Artificial neural networks mirror biological ones, using nodes rather than neurons but building the same sorts of complex interconnections between them. Rather than storing all data in a huge pool to be analyzed as a whole, neural networks remember associations between concepts, streamlining the process of retrieval and analysis. They allow computer scientists to make algorithms for something called “deep learning,” which arranges ideas as layers of definitions. Small concepts collectively define larger ones, which define larger ones, and so on. With enough input information, a sufficiently detailed neural network can learn quite deeply indeed — and, with the possible exception of Google, nobody has access to more raw information than Facebook.
The primary goal of all this is supposedly to improve the venerable News Feed, but when it comes to Facebook the primary goal is always ad sales. Still, powerful deep learning algorithms have the potential to change most of how we interact with social media. What if Facebook or Twitter could recommend a slight rewording to your latest update — switch the word “CPU” for “processor” and get an average 2.4% more attention! What if an algorithm could tell you which cover image will get your photo album the most interest, or search your images for only happy situations? (The “find bikini pics” option will likely be third party, but quick to appear.)
Google’s work with machine learning let an algorithm autonomously learn to recognize a cat’s face.
This isn’t all speculation, either. Google famously taught a neural network torecognize human faces, and Microsoft is using them to bring speech recognition and translation into real time. This all requires that the network make sense of uncategorized information — in other words, it has to be able to turn an arcane posting like “i <3 u babe” into a series of machine learning events, from an increase in babe’s visibility on your News Feed to an automatic alert should the babe in question change their relationship status. Discussing podcasts with a buddy should flag you as a fan of not just the specific shows you mention, but of the medium as a whole, and add to the probability that you also like, say, video games. Deep learning is about making data analysis sophisticated enough to derive your personality from your natural social output.
The AI team tasked with achieving those sorts of gains only recently began this effort, but it brings together experts from all over the field. Yaniv Taigman was the cofounder of facial recognition company Face.com and now works with the team alongside academics like Marc’Aurelio Ranzato and Facebook old-timers like Keith Adams. Though other companies have a head-start, the sheer breadth of information available to Facebook gives its efforts some uniquely personal implications.
Myself, I don’t mind if Facebook peeks in on my activity, just a bit. If we take it as given that social media will be ad-driven for the foreseeable future, we might as well try to make sure those ads remain relevant to our interests. If a banner ad can alert me to a great sale on my impending purchase, or a local tour date for my favorite band, advertising can actually enhance the usefulness of the site as a whole. Sometimes we’re nice enough to place that sort of information in an easily analyzed form — I put Star Trek on my list of favorite TV shows, and an algorithm shows me an ad for Into Darkness. It’s a fairly straightforward process and (leaving aside any possible leakage to government overseers) one that should not overly bother most users.
No comments:
Post a Comment