In 1967, more than a year after his move to Austin, Woody took on one last assignment that involved recognizing patterns in the human face. The purpose of the experiment was to help law enforcement agencies quickly sift through databases of mug shots and portraits, looking for matches.
As before, funding for the project appears to have come from the US government. A 1967 document declassified by the CIA in 2005 mentions an “external contract” for a facialrecognition system that would reduce search time by a hundredfold. This time, records suggest, the money came through an individual acting as an intermediary; in an email, the apparent intermediary declined to comment.
The machine, they concluded, “dominates” the humans.
Woody’s main collaborator on the project was Peter Hart, a research engineer in the Applied Physics Laboratory at the Stanford Research Institute. (Now known as SRI International, the institute split from Stanford University in 1970 because its heavy reliance on military funding had become so controversial on campus.) Woody and Hart began with a database of around 800 images – two newsprint-quality photos each of about “400 adult male caucasians,” varying in age and head rotation. (I did not see images of women or people of color, or references to them, in any of Woody’s facial-recognition studies.) Using the RAND tablet, they recorded 46 coordinates per photo, including five on each ear, seven on the nose, and four on each eyebrow. Building on Woody’s earlier experience at normalizing variations in images, they used a mathematical equation to rotate each head into a forward-looking position. Then, to account for differences in scale, they enlarged or reduced each image to a standard size, with the distance between the pupils as their anchor metric.
The computer’s task was to memorize one version of each face and use it to identify the other. Woody and Hart offered the machine one of two shortcuts. With the first, known as group matching, the computer would divide the face into features – left eyebrow, right ear, and so on – and compare the relative distances between them. The second approach relied on Bayesian decision theory; it used 22 measurements to make an educated guess about the whole.
In the end, the two programs handled the task about equally well. More important, they blew their human competitors out of the water. When Woody and Hart asked three people to cross-match subsets of 100 faces, even the fastest one took six hours to finish. The CDC 3800 computer completed a similar task in about three minutes, reaching a hundredfold reduction in time. The humans were better at coping with head rotation and poor photographic quality, Woody and Hart acknowledged, but the computer was “vastly superior” at tolerating the differences caused by aging. Overall, they concluded, the machine “dominates” or “very nearly dominates” the humans.
This was the greatest success Woody ever had with his facial-recognition research. It was also the last paper he would write on the subject. The paper was never made public—for “government reasons,” Hart says – which both men lamented. In 1970, two years after the collaboration with Hart ended, a roboticist named Michael Kassler alerted Woody to a facial-recognition study that Leon Harmon at Bell Labs was planning. “I’m irked that this second rate study will now be published and appear to be the best man-machine system available,” Woody replied. “It sounds to me like Leon, if he works hard, will be almost 10 years behind us by 1975.” He must have been frustrated when Harmon’s research made the cover of Scientific American a few years later, while his own, more advanced work was essentially kept in a vault.
In the ensuing decades, Woody won awards for his contributions to automated reasoning and served for a year as president of the Association for the Advancement of Artificial Intelligence. But his work in facial recognition would go largely unrecognized and be all but forgotten, while others picked up the mantle.
In 1973 a Japanese computer scientist named Takeo Kanade made a major leap in facial-recognition technology. Using what was then a very rare commodity – a database of 850 digitized photographs, taken mostly during the 1970 World’s Fair in Suita, Japan – Kanade developed a program that could extract facial features such as the nose, mouth, and eyes without human input. Kanade had finally managed Woody’s dream of eliminating the man from the man-machine system.
Woody did dredge up his expertise in facial recognition on one or two occasions over the years. In 1982 he was hired as an expert witness in a criminal case in California. An alleged member of the Mexican mafia was accused of committing a series of robberies in Contra Costa County. The prosecutor had several pieces of evidence, including surveillance footage of a man with a beard, sunglasses, a winter hat, and long curly hair. But mug shots of the accused showed a clean-shaven man with short hair. Woody went back to his Panoramic research to measure the bank robber’s face and compare it to the pictures of the accused. Much to the defense attorney’s pleasure, Woody found that the faces were likely of two different people because the noses differed in width. “It just didn’t fit,” he said. Though the man still went to prison, he was acquitted on the four counts that were related to Woody’s testimony.
Only in the past 10 years or so has facial recognition started to become capable of dealing with real-world imperfection, says Anil K. Jain, a computer scientist at Michigan State University and coeditor of Handbook of Face Recognition. Nearly all of the obstacles that Woody encountered, in fact, have fallen away. For one thing, there’s now an inexhaustible supply of digitized imagery. “You can crawl social media and get as many faces as you want,” Jain says. And thanks to advances in machine learning, storage capacity, and processing power, computers are effectively self-teaching. Given a few rudimentary rules, they can parse reams and reams of data, figuring out how to pattern-match virtually anything, from a human face to a bag of chips – no RAND tablet or Bertillon measurements necessary.
Even given how far facial recognition has come since the mid-1960s, Woody defined many of the problems that the field still sets out to solve. His process of normalizing the variability of facial position, for instance, remains part of the picture. To make facial recognition more accurate, says Jain, deep networks today often realign a face to a forward posture, using landmarks on the face to extrapolate a new position. And though today’s deep-learning-based systems aren’t told by a human programmer to identify noses and eyebrows explicitly, Woody’s turn in that direction in 1965 set the course of the field for decades. “The first 40 years were dominated by this feature-based method,” says Kanade, now a professor at Carnegie Mellon’s Robotics Institute. Now, in a way, the field has returned to something like Woody’s earliest attempts at unriddling the human face, when he used a variation on the n-tuple method to find patterns of similarity in a giant field of data points. As complex as facial-recognition systems have become, says Jain, they are really just creating similarity scores for a pair of images and seeing how they compare.
But perhaps most importantly, Woody’s work set an ethical tone for research on facial recognition that has been enduring and problematic. Unlike other world-changing technologies whose apocalyptic capabilities became apparent only after years in the wild – see: social media, YouTube, quadcopter drones – the potential abuses of facial-recognition technology were apparent almost from its birth at Panoramic. Many of the biases that we may write off as being relics of Woody’s time – the sample sets skewed almost entirely toward white men; the seemingly blithe trust in government authority; the temptation to use facial recognition to discriminate between races – continue to dog the technology today.
Last year, a test of Amazon’s Rekognition software misidentified 28 NFL players as criminals. Days later, the ACLU sued the US Justice Department, the FBI, and the DEA to get information on their use of facial-recognition technology produced by Amazon, Microsoft, and other companies. A 2019 report from the National Institute of Standards and Technology, which tested code from more than 50 developers of facial-recognition software, found that white males are falsely matched with mug shots less frequently than other groups. In 2018, a pair of academics wrote a broadside against the field: “We believe facial recognition technology is the most uniquely dangerous surveillance mechanism ever invented.”
In the spring of 1993, nerve degeneration from ALS began causing Woody’s speech to slur. According to a long tribute written after his death, he continued to teach at UT until his speech became unintelligible, and he kept up his research on automated reasoning until he could no longer hold a pen. “Always the scientist,” wrote the authors, “Woody made tapes of his speech so that he could chronicle the progress of the disease.” He died on October 4, 1995. His obituary in the Austin American-Statesman made no mention of his work on facial recognition. In the picture that ran alongside it, a white-haired Woody stares directly at the camera, a big smile spread across his face.
Source: WiredRelated posts: