By Jérôme Malzac, Innovation Officer, Micropole Group
A fierce debate has gained momentum in recent months and even more so in recent weeks: the use of facial recognition and, by amalgam, that of computer vision, although the subject has been debated in France since 2019 with the announcement of the launch of ALICEM "Authentification en Ligne Certifiée sur Mobile", a new application that allows users to log in via facial recognition to various public services. It is worth noting that France should have become the first European country to implement facial recognition to access online administrative portals.
Long before the decree authorizing the creation and first tests of ALICEM last year, and all the twists and turns that followed (the Quadrature du Net took a position, an appeal was filed with the Conseil d'État, etc.), the use and experimentation of facial recognition in France had already caused a lot of ink to flow.
But what about this facial recognition technology?
If, a few years ago, the results and the error rates were questionable, in particular because the learning bases were not sufficiently differentiated (see MIT studies on the error rates between people of color and white people using IBM algorithms), computer vision algorithms (video and image analysis by computer) have since made enormous progress. Thanks to the development of computing power and the exponential growth of training data (photos and videos of faces in the case of facial recognition), learning is increasingly fast and reliable.
GAFAM has been training AIs to recognize faces for a long time. On the other hand, following the ban in 2019 and for 3 years on the use of facial recognition on the onboard cameras of California police officers (following a demonstration organized by the civil rights association ACLU during which 26 congressmen had been "recognized" erroneously among the faces in a database of offenders via the police solution based on Amazon's "Rekognition") and, More recently, following protests against racism and police violence in the United States related to the death of George Floyd, these companies seem to be backing off, or at least slowing down, on the subject of facial recognition.
Thus, in recent days, Microsoft, IBM, Amazon and Google have had to clarify their position on the issue and are calling for a legislative framework for this technology, each with its own announcement:
Microsoft, via its president and general counsel Brad Smith, wants to deny access to its technology to U.S. police forces "until there is strong federal legislation based on human rights".
IBM announces, via its new CEO Arvind Krishna, to withdraw from the facial recognition market.
Amazon acknowledges California's decision by announcing a one-year moratorium on its Rekognition software to give the US Congress time to "put in place appropriate rules".
Or the CEO of Google, who announced last January in Brussels that he wanted a A "judicious regulation". European Union on artificial intelligence (AI) and safeguards around facial recognition.
However, GAFAM are not the only ones to have launched themselves on the subject: many manufacturers in the field of telephony, private security and video surveillance companies and even governments have plunged into the development of facial recognition solutions.
So what about the use of this technology?
Let's first talk about the most iconic and, in my opinion, the most advanced player of all: China. More than 200 million cameras in China (450 million are planned by the end of 2020) are filming the population by continuously running facial recognition algorithms.
If we look at the good side of the technology, we can see several benefits:
Access to student housing secured by facial recognition.
Faster payment solutions without the risk of forgetting the means of payment that you can't lose while it's on your shoulders.
Shorter waiting times when it comes to queuing for access to an event, etc.
Between those who see them as beneficial and those who see them as an infringement of freedoms, certain uses are discussed, including:
Quickly find wanted persons in a crowd: a few months ago, a man wanted by the police was identified and arrested in a concert among 50 000 other spectators.
Identify and track down individuals harming property or people: more than 25 people were identified and arrested in 2018 following criminal acts thanks to facial recognition via glasses worn by police officers.
If we place ourselves on the dark side of the force, we can see the failings that such technology allows in an "authoritarian" state:
Thefine received at home, if you don't cross in the right lane or when the little man is green...
Social credit, a rating system based on your actions that gives you access to certain services, schools, jobs, or prohibits you from selling tickets, access to certain services ... The episode of Black Mirror on social rating is no longer a fiction!
What about the deployment of this technology in France?
At the moment, France has a little more than one million surveillance cameras, which is not to be ashamed of if you calculate the ratio between the number of cameras and the total population compared to the Chinese population which has 200 million cameras.
In 2019, the City of Nice had launched a facial recognition experiment during its annual carnival. This one had been authorized by the CNIL since it consisted in asking the authorization to 5,000 volunteers to share the time of the experiment the photo data of their face to test the ability to be found in the crowd. If this experimentation was a success from a technical point of view (more than 98% of positive results obtained), it was a little less so from the point of view of acceptance, since many detractors cried out about the risk of a deprivation of our individual liberties.
The PARAFE system, which links your face to the data in your biometric passport, is already used in some airports (Nice, Orly, Roissy, Marseille, etc.) for automated border control and is being extended to check-in and boarding (initially for Air France flights). This will save time and reduce the time needed to board the plane: in less than a second, the system will know if you are authorized to board or not. This system will become essential in the future, if the number of flights and rotations per day continues to increase.
What are the benefits of using these technologies?
Let's approach facial recognition from a "rule of law" perspective, including the contributions of computer vision, image analysis and facial recognition as a whole.
Here are some very positive real-world applications that have been made possible by this technology:
Following an experiment by the New Delhi Police, 3,000 lost children were reunited with their families in 4 days thanks to facial recognition.
Part of the US Civil War archive has been reconstructed using the people in the photos of the time and facial recognition.
The ease of unlocking smartphones by face is constantly progressing, even if the beginnings were complicated with hack solutions, such as a simple photo rather than a real face. In my opinion, we should bet on the complementarity of technologies for a strong authentication: facial recognition backed up by pulse identification and infrared analysis to verify that it is a face and not a simple photo or a 3D scan, for example.
Facial recognition coupled with the Traitement des Antécours Judiciaires (TAJ) file, which gathers information on police and gendarmerie procedures, contains 18.9 million records of suspects, nearly 8 million photos and represents 6 terabytes of data, has helped solve crucial investigations such as the knife attack that took place in Paris in May 2018 or the "bicycle terrorist" who planted a bomb in Lyon in May 2019.
More generally, image analysis can be useful in many cases. As soon as we know how to model the normality of a situation, we can detect abnormal situations.
As part of the development of the smart city, video surveillance allows various contributions, including:
Detect traffic problems or accidents in order to accelerate the dispatch of help or to anticipate dangerous slowdowns or problems before accidents occur.
Detect a change in behavior (a crowd suddenly running, people on the ground, beginning of a fight, abandoned luggage, etc.) that would trigger an alert about a conflict situation.
As we approach important events such as the 2023 Rugby World Cup or the 2024 Olympic Games in France, it seems to me that it would be appropriate to test and make robust this type of anticipation and reaction device, to which image or sound analysis and facial recognition have much to contribute.
In the medical field, facial recognition offers many possibilities:
By a simple emotional analysis of the face, it allows to detect and evaluate the level of pain of the patient, which is sometimes not translated by the words to the height of the real trauma.
To complete and accelerate diagnoses and thus anticipate certain diseases through the analysis of radio, MRI, CT scans ... especially in the fight against cancer with significant progress in recent years thanks to medical AI.
In the retail field, these technologies contribute to the improvement of the customer experience in store:
Theanalysis of customer flows via security cameras allows for the optimization of the layout of the shelves or the traffic passages in the stores.
VIP customer recognition offers personalized support.
Video analysis allows you toskip the checkoutline and pay with facial recognition.
Over the past few months, we have seen how COVID-19 has changed our lives and our habits with the mandatory wearing of masks in certain transportation or daily situations, and the respect of "social distancing" in certain circumstances. What if these measures became the norm in the event that this virus were to persist over time? Without going as far as identifying the person, because I am not in favor of punishment but rather of respect and collective awareness, computer vision, image analysis and facial recognition with a view to counting, would help in the fight against the epidemic. This could be done by regulating the flow of people in a store or a public place via counting and/or analysis of the distance between them, for example, or by adapting in real time the circulation of people and distances according to the detection of the mask on faces.
Nevertheless, the CNIL risks slowing down the experiments by expressing reservations and doubts about the use of thermal cameras capable of evaluating temperature or the wearing of a mask. On Twitter, the CNIL mentioned the risks of the "trivialization of intrusive technologies", and of "increased surveillance, likely to undermine the proper functioning of our democratic society". Here is a position and words that go beyond the simple request for vigilance but which has the merit of encouraging to think about a concrete framework of use.
In conclusion, I think that we should not be systematically pessimistic about technological progress, even if we should not allow everything. It is certain that no technology, no solution, is and will never be 100% infallible. Besides, man would never have set foot on the Moon if it had been necessary to reach 0% risk before sending Saturn V and Eagle into space with automatic pilots on computers that have a computing power lower than our current smartphones. The only way to guarantee a level of safety and an appropriate use of a new technology is to let experimentation and research of the benefits brought VS the risks incurred. I am talking about an experimentation known and accepted without constraint by its user and framed in the respect of the private life.
In the case of video and photo analysis, with technological progress, improvements in algorithms and learning databases, false positives will decrease to reach a reliability rate comparable to human error. Even if some experiments, applications or ways of doing things are questionable for some (for example, Google has been criticized for paying volunteers with vouchers in exchange for a photo to enrich the learning database), image analysis technology cannot be ignored or rejected.
If I am really not one of those who like to create laws and regulations to excess, I believe that we are already constrained by many regulations and that we will certainly have to find effective ways to frame and regulate this technology. I remain convinced that it has a lot to offer us, that its future depends on its use by the rule of law.
The problem is not the technology, but the uses by man!
Is technological innovation a source of progress, allowing us to go from the mastery of fire to today's AI and computer vision technologies in 400,000 years, or a real threat to our free will and individual liberties? For my part, my choice is made!