The Pros and Cons of Facial Recognition Software
The meteoric rise of open-source facial recognition and GPU-based machine learning repositories has progressed faster than any legal restrictions or ethical, and policy guidelines could have been addressed in the past five years. Estimated at $3.4 billion in 2019, the global market is projected to grow to more than $10 billion by 2027.1
Biometrics, security applications, marketing, and attendance systems are key sectors that have fueled the leader in the computer vision software business over the last five years, all backed by recent developments in machine learning and the expanding usage of lightweight mobile deep learning frameworks.
How businesses are currently using facial recognition applications
The best-known uses of facial recognition software are commercially available products from companies that have invested the most in such systems, each with a large user base to receive and use a considerable amount of incoming facial identification data.
Apple Face ID
Apple’s sophisticated face-based authentication system for iOS uses several aspects of facial scanning to verify the user, including 30,000 individual points of facial data and infrared imaging to determine “liveliness.”
In June 2020, Apple announced that it would expand the scope of Face ID with its web authentication API, allowing users to log into websites using their faces. The technology community generally welcomes this move, given that Apple has no corporate authority to commercialize user data.
Facial recognition has been a cornerstone of Facebook’s AI research for the past decade. Since 2016, after publishing its DeepFace neural network FR system, the social network has used it to identify users tagged in photos and videos uploaded by others.
In September 2019, under pressure from lawsuits, privacy campaigns, and the federal government, Facebook gave its users the option to disable several facial recognition settings for their accounts.10 In January 2020, the company also agreed to pay $550 million to settle a lawsuit over facial recognition data breaches in Illinois.
Google is one of the world’s most funded AI researchers, and much of that effort is devoted to improving facial recognition architectures. The company is applying its research to photo tagging (in Google Photos and for search capabilities) and Android-based login, among other applications.
Google, one of the defendants in an Illinois lawsuit, alleges, along with Microsoft, Facebook, and Amazon, that the search giant abused user-uploaded content in violation of state privacy laws. It’s not just a lawsuit, as more U.S. states are grappling with the tech giants’ shameless use of user-generated content for facial recognition purposes.
In October 2019, Google temporarily suspended facial recognition research for its Pixel 4 smartphone after it was revealed that contract researchers selected minority students and homeless people as test subjects to rebalance the racial biases inherent in FR systems. Work on the system has since resumed.16
Amazon
Amazon’s Rekognition system was first leased to police agencies in 2017. Further deployments of the system, including the controversial and unsuccessful use by Orlando police in Florida in 2019, along with other incidents, culminated in a massive protest by Amazon employees, who, along with numerous voices of scientists, successfully lobbied their superiors to restrict the sale of facial recognition products to police, military and government agencies.
In June 2020, Amazon surprised the industry by refusing to sell its facial recognition technology to police agencies, saying it would end a one-year moratorium pending new legislation from the U.S. Congress.
The changing face of facial recognition
There is plenty of evidence that the “wild west” years of facial recognition are coming to an end under pressure from privacy advocates and politicians who have to reconcile the state’s interests with the will of the voters.
Over the next, five to ten years, increased regulation should bring facial recognition technology research and adoption to a more stable state. The new legislative boundaries are likely to weed out the early opportunists. Still, they will leave behind a regulated industry ready to enter a more traditional business cycle of competition and conglomeration.
In addition to the fact that the advent of COVID-19 has changed the landscape of facial recognition with intelligent video analytics, other political and social factors, both private and public, have recently begun to pressure its use:
In June 2020, IBM sent a letter to the U.S. House of Representatives saying it would no longer supply facial recognition and analysis software. It called for a “national dialogue about whether and how facial recognition technology should be used by domestic law enforcement.”
The tentative introduction of a facial recognition scheme by the U.K. Metropolitan Police and the controversy surrounding the unsystematic use of facial recognition has reinforced calls for additional oversight mechanisms.
At the same time, as prominent privacy regulators warn that AI-based facial recognition could become illegal in the European Union, pan-European police forces are pushing for a pan-European facial recognition database.
The lack of consensus creates a turbulent and volatile environment for business investors in facial recognition. When even a well-known company such as IBM, under pressure from shareholders responding to public sentiment, rejects the technology as a PR commitment, one can assume that automated facial recognition technology has reached a tipping point.
Regulation as a business facilitator of facial recognition systems
Increased regulation and rising public mistrust are probable indicators that new technology is establishing a long-term partnership with society, negotiating the parameters of its future success. Some of the most revolutionary technical advancements, such as the printing press and genetic engineering, have been regulated and rationalized, often to the chagrin of its proponents.
When governments throughout the world regard crowdsourced data and open-source technology as a global threat, the consequence has typically been absorption and control rather than annihilation or ban.
companies, are a massive help in developing facial recognition projects. The key to project success is management and due diligence regarding project scope and legal protection.
Using face recognition to protect the future
While legislation governing facial recognition technology changes on a regular basis, both globally and in individual U.S. states, a strict approach to data management and thorough familiarity with local and national laws are critical to the long-term viability of a business facial recognition deployment.
Be aware of existing and anticipated regulations
Consult with national and, if applicable, state legislators on privacy and data management issues related to facial recognition technology. Be mindful of pending bills and amendments that may change regulations in the future.
Research current examples of facial recognition technology implementation in the private and public sector and familiarize yourself with recent legislation that promotes its performance. These could be projects in the interest of national security or where the use of facial recognition is officially recognized as uncontested, such as in workplaces where facial recognition is included in employees’ limited right to privacy; in prisons and other public institutions; in educational settings; in experimental scenarios where subjects have specifically agreed to face recognition and where local and national law permits such exceptions.
Build data governance mechanisms into your facial recognition projects from the start
Even if not required by current regulations, your facial recognition project should have accessible, human-readable data retention policies, which should meet at least the minimum requirements of applicable laws in your area.
Provide mechanisms by which users can be made aware of the facial recognition data they receive, as well as a means by which users can delete their data and/or opt-out of the facial recognition scheme. Even if there is currently no legal need to provide such functionality, it may be needed in the future.
Keep detailed long-term logs
Implement a comprehensive and secure logging system by applicable laws that may specify policies regarding the scope, detail, and retention of logs for the use and transmission of facial recognition data.
Define and publish clear information about the sharing of facial recognition data
A transparent approach to sharing user facial recognition information is critical. Except for court requests from authorities or when data retention requirements may expire, it should remain possible to provide a full accounting of where and when facial recognition data was shared with the parties included in the opt-out terms or generally following applicable local and national law.
When facial recognition is a subset of object recognition/segmentation, PR nightmares can multiply: Google was put in an awkward position after its facial recognition program identified two ethnic subjects as “gorillas.”
It is inevitable that the facial recognition program’s goals will be brought down to “average” subjects, who may or may not share many common characteristics. But it is essential to ensure that your tools and methods for creating datasets are unbiased. Don’t strive for results that you already expect, and make sure that all possible distilled data have been designed neutrally, being prepared to demonstrate this later if necessary.
Final Thoughts
If identity has become the new currency, the development of facial recognition in the next ten years will require as much diligence and compliance as the creation of a banking system, where dealing with the money itself is relatively trivial logistics, but controlling its safe and lawful movement and use is a major challenge.
These business models have failed to understand that a particular deployment of facial recognition will inevitably run up against a firewall of public objections, leaving the sector to develop into a regulated and practical industry with acceptable checks and balances and appropriate entry conditions.
This serious approach will distinguish the most successful use of facial recognition in the public and private sectors in the coming years.
