Facial recognition, between benefits and risks to individual freedom

By Jérôme Malzac, Innovation Officer, Groupe Talan

A fierce debate has gained momentum over recent months and even more so in recent weeks: the use of facial recognition and, by extension, computer vision — although the topic has been debated in France since 2019 with the announcement of the launch of ALICEM ("Authentification en Ligne Certifiée sur Mobile"), a new application enabling users to log in to various public services via facial recognition. It is worth noting that France was set to become the first European country to implement facial recognition for accessing online administrative portals.

Long before the decree authorizing the creation and first tests of ALICEM, and all the developments that followed (the position taken by La Quadrature du Net, the appeal lodged before the Conseil d'État, etc.), the use and experimentation of facial recognition in France had already generated considerable debate.

But what is the reality behind this facial recognition technology?

While a few years ago the results and error rates were questionable — particularly due to insufficiently diverse training datasets (see MIT studies on error rates between people of color and white people using IBM algorithms) — computer vision algorithms (computer-based video and image analysis) 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), training processes are becoming increasingly fast and reliable.

Big Tech companies have therefore been training AIs to recognize faces for a very long time. However, following the 2019 ban — lasting three years — on the use of facial recognition on body cameras worn by California police officers (following a demonstration organized by the civil liberties organization ACLU, during which 26 lawmakers were incorrectly "identified" as appearing in a database of offenders using a police solution based on Amazon's "Rekognition"), and more recently in the wake of protests against racism and police violence in the United States following the death of George Floyd, these companies appear to be stepping back, or at least slowing down on the issue of facial recognition.

In recent days, Microsoft, IBM, Amazon and Google have each been called upon to clarify their position on the matter, each making their own announcement:

Microsoft, through its President and Chief Legal Officer Brad Smith, wants to deny access to its technology to American police forces "until there is strong federal legislation grounded in human rights."

IBM, through its new CEO Arvind Krishna, announces its withdrawal from the facial recognition market.

Amazon endorses 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."

Google's CEO had also announced, last January in Brussels, his desire for "sensible regulation" from the European Union on artificial intelligence and safeguards around facial recognition.

Yet Big Tech is not the only player to have ventured into this space: numerous manufacturers in the telecommunications sector, private security and video surveillance companies, and even states have immersed themselves in the development of facial recognition solutions.

What are the advantages of using these technologies?

Let us approach facial recognition within the framework of a "rule of law," encompassing the contributions of computer vision, image analysis and facial recognition as a whole.

Here are some concrete, highly positive applications that have been made possible by this technology:

Following an experiment by the New Delhi police, 3,000 missing children were reunited with their families within 4 days thanks to facial recognition.

Part of the archives of the American Civil War was reconstructed through the people appearing in photographs of the era and facial recognition.

The ease of unlocking smartphones by face continues to improve, even though early attempts were fraught with hacking workarounds, such as using a simple photograph rather than a real face. In my view, we should bet on the complementarity of technologies for strong authentication: facial recognition backed by pulse identification and infrared analysis to verify that it is a real face and not simply a photo or a 3D scan, for example.

Facial recognition coupled with the French Judicial Record Processing database (TAJ), which gathers information on police and gendarmerie procedures, contains 18.9 million records of persons implicated, nearly 8 million photos and represents 6 terabytes of data. This has made it possible to solve crucial investigations, such as the knife attack that took place in Paris in May 2018 or the "cyclist terrorist" who planted a bomb in Lyon in May 2019.

More broadly, image analysis can be useful in many cases. From the moment one is able to model the normality of a situation, it becomes possible to detect abnormal situations.

In the context of smart city development, video surveillance offers various contributions, including:

Detecting traffic problems or accidents in order to accelerate the dispatch of emergency services or to anticipate dangerous slowdowns and issues upstream of accidents.

Detecting a change in behavior (a crowd suddenly running, people on the ground, the start of a fight, an abandoned bag, etc.) that would trigger an alert about a conflict situation.

With major events such as the 2023 Rugby World Cup or the 2024 Olympic Games in France approaching, it would seem timely to test and make robust these types of anticipation and response systems, in which image or sound analysis and facial recognition have much to contribute.

In the medical field, facial recognition offers numerous possibilities:

Through simple emotional facial analysis, it enables the detection and assessment of a patient's pain level, which is sometimes not expressed in words commensurate with the actual trauma.

To supplement and accelerate diagnoses, and thus anticipate certain diseases through the analysis of X-rays, MRIs, scans, etc. — particularly in the fight against cancer, where significant progress has been made in recent years thanks to medical AIs.

In the retail sector, these technologies contribute to improving the in-store customer experience:

• Customer flow analysis via security cameras enables the optimization of shelf layouts and circulation pathways within stores.

• Recognition of VIP customers enables personalized service.

• Video analysis makes it possible to avoid queuing at the checkout and to pay using facial recognition.

Over recent months, we have witnessed how profoundly COVID-19 has disrupted our lives and habits, with the mandatory wearing of masks in certain forms of transport and daily situations, and the observance of "social distancing" in certain circumstances. What if these measures were to become the norm should this virus persist over time? Without going as far as identifying individuals — as I am not in favor of sanctions but rather of respect and collective awareness — computer vision, image analysis and facial recognition used for counting purposes could help in the fight against the epidemic. This could take the form of regulating the flow of people in a store or public place through counting and/or analyzing the distance between them, for example, or the real-time adjustment of people's circulation and distances based on the detection of masks on faces.

Nevertheless, the CNIL (French data protection authority) is likely to put a brake on experimentation by raising reservations and doubts about the use of thermal cameras capable of assessing temperature or mask-wearing. On Twitter, the CNIL raised concerns about the "normalization of intrusive technologies" and "increased surveillance, likely to undermine the proper functioning of our democratic society." This is a position and a set of words that go beyond a simple call for vigilance, but which has the merit of prompting reflection on a concrete framework for use.

In conclusion, I believe we should not be systematically pessimistic about technological progress, even if we should not permit everything either. It is certain that no technology, no solution, is or ever will be 100% infallible. Besides, mankind would never have set foot on the Moon if a 0% risk threshold had been required before sending Saturn V and Eagle into space with autopilots running on computers with less computing power than our current smartphones. The only way to guarantee an appropriate level of security and appropriate use of a new technology is to allow free rein to experimentation and the exploration of benefits versus risks. I am referring here to experimentation that is known and freely accepted by its user, and conducted in compliance with respect for privacy.

In the case of video and photo analysis, with technological advances, improvements to algorithms and training databases, false positives will decrease to reach a reliability rate comparable to that of human error. Even if certain experiments, applications or practices are questionable to some (for example, Google has been criticized for compensating volunteers with gift vouchers in exchange for a photo to enrich its training database), one cannot ignore or reject image analysis technology.

While I am genuinely not among those who enjoy creating laws and regulations in excess, I believe that we are already constrained by numerous regulations and that effective means will certainly need to be found to frame and regulate this technology. I remain convinced that it has much to offer us, and that its future depends on its use within the rule of law.

The problem is not the technology, but the uses made of it by human beings!

Is technological innovation a source of progress — enabling us to move from the mastery of fire to today's AI and computer vision technologies in 400,000 years — or a genuine threat to our free will and individual freedoms? As for me, my mind is made up!