9 strange discoveries about facial recognition technology

The development of facial recognition technology has greatly accelerated. It has been 43 years since researchers have been working on the subject. One recent study presents historical results compiled from more than 100 data sets used to shape systems specializing in facial recognition. The results are surprising.

How does facial recognition technology work? What are the privacy issues? What interesting facts are related to this topic? As facial recognition systems improve, is society losing its privacy?

What are the conclusions of 43 years of research on face recognition?

1.Technology is proving to be unreliable. Example? The first attempts to record and identify the faces of motorists speeding along the highway near the Robert F. Kennedy Bridge in New York City in 2018 were unsuccessful. Interestingly, the system worked flawlessly under academic conditions.

2.Research on facial recognition began in universities. However, the Department of Defense (DoD) is responsible for developing this technology and The National Institute of Standards and Technology (NIST), which in 1996 allocated $6.5 million to create the largest dataset to date. The government became interested in this area of work due to the fact that it did not require active citizen participation, unlike obtaining fingerprints, for example.

3.At the very beginning of the development of facial recognition technology, portrait photos were used. However, these were not very accurate, so the results could be misleading.

4.Further work on facial recognition involved searching for people on Google. This happened when researchers wanted to expand the datasets and go beyond person portraits. It became increasingly difficult for researchers to manage such diverse data, which involved various processes, such as obtaining the consent of the person being photographed, recording demographic arrangements, maintaining the quality of the dataset, and so on.

A 2007 dataset called Labeled Faces in the Wild has taken on Google, Flickr, YouTube and other online photo banks. This allows users to find more photos in them, but privacy rights are somewhat diluted.

5 Another breakthrough in facial recognition technology has come from Facebook. The social network revealed that it created its own DeepFace database in 2014. DeepFace identifies human faces in digital images, and uses a nine-layer neural network with more than 120 million connections. The system has been tested on four million images uploaded by Facebook users using deep learning technology.

6 The groundbreaking technology used by Facebook, however, breached the data of too many people. The site’s creators paid a fine for using photos uploaded by Facebook users to enable facial recognition (without obtaining users’ consent).

The DeepFace project was not announced in the press, and the way it worked was “tag suggestions.” The idea was to suggest the person in the photo the user wanted to tag. This would have helped the development of the system, but it caused further data breaches, so “tag suggestions” were eventually abandoned.

7 Facial recognition technology has used the images of 17.7 million people, and these are only public datasets. In fact, the number and identity of people who have become unwitting participants in the development of facial recognition technology is unknown.

8 The automation of facial recognition has led to unequal representation and the sticking of “labels” on people. Systems have evolved beyond identifying a face or person to labeling features in offensive ways, such as by using the phrases “pudgy,” “pale skin,” “pointy nose,” “bags under the eyes,” etc.

What’s more, faces considered “Western” have become the default, such as in training sets. According to the study, discrimination in artificial intelligence can reinforce discrimination in the real world.

9 Today, facial recognition technology is widely used, from advertising targeting to government surveillance. Amazon Rekognition technology and related data have been diverted to the police. On the other hand, facial recognition systems can significantly improve programs to analyze shopper sentiment and better understand the needs of potential customers.