In late January, Amazon was once again at the center of a scandalinvolving its artificial intelligence for face recognition. it became Clear that its system Rekognition is not able to distinguish the floors are black, and therefore inferior in features counterparts from Microsoft and other companies. The logical solution would be to use higher-quality training data, but researchers from the artificial intelligence lab at the University of Massachusetts came up with a more effective way.
It became known that they develop an algorithm that automatically saves sets of training data from photos, which can lead to error in recognizing people with dark skin. The researchers do not promise that this method will give 100% result, but the number of errors will obviously be less. This is evidenced by internal tests scientists: artificial intelligence was 60% less biased, and it has not lost the recognition accuracy. The technology will also reduce training time — the use of large amounts of data slows down the process.
Absence of bias — a very important characteristic for artificial intelligence. It is increasingly used by the police, so even a small error of the algorithm may be the reason that the prison can be planted completely innocent people. Many people do not like the widespread use of face recognition system, but if they are to be used — it is important to work as accurately as possible.
They are already being used even in Russia. For example, during the world Cup 2018 face detection algorithm was implemented in the Moscow metro station “Frunzenskaya”. Thanks to her, the police were able to quickly discover a manwho is wanted for robbery. He became the first criminal who was caught using face detection. Its cost, by the way, is estimated at 4 billion rubles.
Want to speak on the subject of face recognition system? Welcome in comments or in our Telegram chat, where hundreds of participants are always happy to talk on the subject of science and technology!