Facial recognition software is widely used to identify people and until recently it was one of the few areas where humans out-performed machines. Until about five years ago, the errors in machine identification of faces was about 25%, i.e. barely reaching 75% accuracy, while humans were in the range of 97.5% accuracy.
However, in the last few years, new developments in software have allowed machines to exceed the human accuracy level, and such software is now utilized by Baidu, Google Facenet, tencent bestimage, Deep ID3, all of which reach over 99.5% accuracy. Facebook’s software is now neck and neck with human accuracy levels and is being provided by the company to assist the Facebook user to keep track of your social network friends who show up in other pictures in other parts of the Internet.
In addition, the capacity for being able to sift out of a very large number of pictures some commonalities and identify common categories (cats, dogs, birds, human, etc.) used to be another area where humans used to best the machines. This is no longer true. Without coaching a “Deep Learning” program was able to sift through hundreds of thousands of pictures and identified cats as a category. While humans succeed on average 95% or the time, computers who had error terms of 25% or more in 2011 now routinely do better than humans.
Field after field seems to fall to the machines’ mounting abilities. So what does the future hold?