The real challenges of facial recognition technology
June 21, 2019 3:15 pm
5 min read
This content has been provided by IDEMIA National Security Solutions (NSS).
Over the past few years, facial recognition technology (FRT) has improved in leaps and bounds. Some developers have gotten FRT programs to make matches with 99% accuracy, while reducing false positives to once in a million events. Of course, there are some that do not work as well, and there are also some that are tuned to work for particular applications. As with most emerging technologies, consumers need to do their homework and often must make tradeoffs.
Facial recognition and video analytics technologies have also been improving thanks to extensive R&D and growing galleries of raw image data from law enforcement and intelligence communities. That data can be fed to FRT engines to give them more experience and improve accuracy, a process which also helps the agencies. In fact, without improved FRT, agencies would continue to be awash in pictures and video that they would be unable to analyze in a timely manner if at all.
There is no question that the need for efficient and accurate FRT is a national security concern. But the greatest challenges may not be technological. Experts like President and CEO of IDEMIA National Security Solutions Scott Swann say the biggest hurdle for the evolving technology is the challenge of public understanding and acceptance. The public has not always been shown the technology in an accurate light, leading to negative perceptions regarding its use and deployment.
One concern often raised is that these systems can exhibit racial or gender bias. Swann said that this broad assertion is misleading. Many of the studies that expose this potential bias are not based on modern systems or the most advanced algorithms. The error rates, in fact, have come down significantly as a result of advances in the technology. Lastly, it’s critical to know that these systems don’t make definitive identifications without a human in the loop – so PR stunts that highlight false matches don’t tell the true story. Instead, these systems provide a gallery of potential matches or a probability of a certain match.
Of course, as with any tool employed by law enforcement or intelligence communities, concerns about privacy will inevitably be raised. Civil liberty advocates rightly demand that individuals receive clarity about how this technology may be used.
Setting up a clear, concise policy concerning the use of FRT will go a long way toward alleviating those concerns, Swann said. According to him, agencies should only use FRT when there is a compelling, lawful and appropriate need, and even then use best practices when doing so. Keeping accurate records such as impact assessments will also help make certain that policies are followed, he added.
Proper training of anyone involved in the FRT process is essential to making sure privacy guardrails are not exceeded. “Facial recognition systems are not autonomous,” said Swann. “They require trained analysts to operate efficiently and interpret search results. If we want to get dependable results, we need to train our analysts in the science of facial comparison and how these systems operate.”
Many of our natural instincts as humans run contrary to accurate FRT analysis, Swann added. “For example, our brains tend to see a face as a distinct unit, rather than focus in on specific and often very small features,” he said. “Only rigorous training and regular proficiency testing can overcome such impulses.”
As an example, the New York Police Department has become increasingly vocal about the appropriate use of FRT. Their Real Time Crime Center (RTCC) has hired and trained facial examiners who often review hundreds of candidate photos when attempting to identify an individual. They use corroborating video or photo evidence with pictures from previous arrests or other databases available to the investigators.
Public acceptance of any investigative tool is always a huge challenge, and FRT is no exception. A lot of this reluctance centers on misconceptions that many people have about how the technology works, and how it is used.
One major cause of these misconceptions is how FRT is represented in movies and television, Swann said. If they were to be believed, someone just pushes a button and sometime later the computer pings like a toaster oven and shows the exact desired results. “Hollywood has been great for publicizing the wow factor of FRT, but it’s critical that we separate entertainment from reality,” explained Swann. We are not living in a dystopian police state where citizens have no voice in the face of pervasive surveillance. That’s not even feasible from a financial, legal or technological standpoint. On the contrary, government and industry leaders are actively engaging broad stakeholders to develop responsible, balanced and lawful guidelines for facial recognition.”
A lot of the public’s reticence will be abated as this technology becomes more of a staple in the consumer arena, Swann said. For example, unlocking one’s phone with FRT is becoming commonplace, and this familiarity will make acceptance of its use in law enforcement and national security easier to understand and digest.
The Time Is Now
While it’s true that there are many obstacles to overcome in order to have FRT gain the public acceptance it needs in order to reach full appropriate implementation, experts like Swann suggest that not doing so is far more dangerous.
“Recent FRT algorithm tests by NIST indicate that China and Russia are gaining against the U.S. and our allies,” Swann explained. Such countries can feed to artificial intelligence systems an ever-increasing network of surveillance sensors and swaths of personal information collected by the government. This greatly accelerates the learning process of those technologies, allowing the host countries to gain a possible advantage.
“We still have some strategic advantages with FRT, but this is a wakeup call,” urged Swann. “Countries with little or no privacy protections make vast repositories available to their facial recognition developers for algorithm improvement. The answer is not to ignore privacy considerations, but U.S. industry must work together with government to find innovative methods for maintaining our strategic edge.