How many billions are the last billion’s biometrics worth?
Developing world identity systems lag behind those in the richest countries, and biometric systems to address the gap face technical and implementation challenges. The human cost of not meeting those challenges is real, but the business case is still being worked out.
Earlier in the maturation of biometrics as a digital technology and an industry, providing identification in the developing world was largely a matter of answering questions with technical solutions. Can biometric sensors be small enough and efficient enough to be transported and used without optimal infrastructure? Can biometrics be captured out of doors in bright sunshine? Can the collected data be integrated into a usable, deduplicated system?
The number of people without legal identification in the world has declined to roughly one billion, and new questions around identification for development are arising, as the old ones are answered. Questions like “how can we afford to reach the most remote people?”, and “how can biometrics be associated with early childhood records?” require innovative types of delivery. Several companies are attempting to address these new questions with different approaches, including innovative hardware designs and support models.
Kidprint is a research project run out of the Center for Health at the University of California at San Diego’s Design Lab, which developed the ION device to provide identification for very young people to enable effective immunization and early childhood health checks. The KidPrint research team has achieved functional true acceptance rate of 99.9 percent enrolling and authenticating children as young as 3 days old with ION, KidPrint project lead Eliah Aronoff-Spencer told Biometric Update in an interview.
“The neat trick here is: You don’t have to use our device for everything, you just have to use it in the early months,” he says.
This is key, as Michigan State University Professor and world-renowned biometrics expert Anil Jain pointed out in an interview with Biometric Update that collecting biometrics from anyone over two years of age is, at this point, a “solved problem,” from a technical perspective. Infant biometrics, however, are still a challenge.
Closely following human centered design principles, starting with “make sure you’re solving the right problem,” the Kidprint team worked with Ken Warman of the Bill & Melinda Gates Foundation on refining the problem statement, and then experimented with a range of existing technology and 3D-printed objects to try different shapes. They took a modular prototyping approach, and generated what Aronoff-Spencer calls a “wall of shame” filled with rejected shapes and designs.
Simprints is another company with a developing-world focus that went through many designs, utilizing human-centered design and other approaches to bridge the gap to what should be possible with the technology, from the less impressive results on the ground. The Children’s Investment Fund Foundation is currently one of Simprints main funding supporters, backing a deworming campaign in Ethiopia. One of its other major current projects is identifying not infants, but expecting mothers. Maternal healthcare workers in Bangladesh are deployed throughout large areas to provide care and advice to numerous women at the same time, Simprints Chief Operating Officer Sebastian Manhart told Biometric Update.
“Without biometrics, without unique ID, its just guesswork. You simply cannot do what is required. But the operational challenges are absolutely massive,” he says.
How to deliver biometric identification to everyone:
Jain states that any discussion about the state of biometrics-backed identity in the developing world has to begin with Aadhaar. The world’s biggest biometric database registered hundreds of millions of people who had previously held no legal identification, with the full enrollment project taking less than a decade from first steps to completion. Whatever issues the Aadhaar program has had in terms of operations and legal footing, it has been an unqualified success as an enrollment effort.
“That’s the most important innovation in the developing world,” Jain argues. “Everybody wants to duplicate it, but nobody has been able to successfully, because execution is the main thing. You can have as much technology as you want, but if you do not properly train the individuals who are collecting the biometric data, and if you do not have the proper standards for collecting the images, then it’s going to fail.”
Fingerprint sensors and algorithms have already advanced to the point where most people can be easily and cost-effectively enrolled, but beyond a certain baseline, their continued improvement add little value for most situations.
“Then there are all the other factors we learned about from experimenting with prototypes and different technologies that have nothing to do with how good you sensing technology is,” explains Simprints Chief Product Officer Alexandra Grigore. “You could have the best sensor technology out there, if people don’t know how to use it properly, it’s useless. Like any technology; garbage in, garbage out.”
Training is one of several things India’s Unique Identification Authority of India (UIDAI) did differently – and better — than most biometric projects, according to Jain. Front line workers have to be trained on how devices work, as well as proper finger placement, the appropriate application of pressure, and when to clean the subject’s fingers. This is particularly true for difficult subjects, like infants.
“There’s a whole sequence of procedures which we developed for capturing the fingerprints of babies,” Jain notes. “If you don’t follow it, the data will be noisy, and no matter what kind of matcher or sensor you have, it’s not going to work.”
Grigore also argues the importance of training. “First you have to have the right technology, and then you have to have the right design, and the right user experience. I think the third most important thing is around training. This is where we at Simprints invest a lot, and that’s why we take a service model approach. We don’t just ship the technology and pray.”
In some situations, Grigore says, decision makers who may have high levels of education and a comfort level with technology believe a device or project will work, but the field workers do not all fully understand the importance of collecting good quality data.
Ion Biometrics is the B-Corp which is the business side of the Kidprint research project, and is led by Michael Kleeman, senior fellow of the UC Institute on Global Conflict and Cooperation. Ion is planning to license its device to partners in developing regions, Kleeman says, in part in order to take advantage of localized distribution and support.
Improved levels of support and training may lead higher rates of successful enrollment and matching, but getting quality fingerprint data in developing contexts is also made challenging by the usual range of infrastructure and budgetary limitations. Even when they are dealt with successfully, fingerprinting may not even be possible for some people.
“When you go out into the field and you work with people who work have, let’s call them ‘real jobs,’ even in developed countries you have a failure rate due to the fingerprints getting rubbed off,” Aronoff-Spencer says. “Then as you get older people still working, that failure rate starts to get pretty significant.”
This is why Aadhaar enrolls iris biometrics as well as fingerprints. For most projects in the developing world, however, “solving the right problem” means precise targeting since, Manhart notes, biometrics are still considered expensive by many governments and international development partners. That is why operational success is so important, both for the implementing body and the people who are the project’s intended beneficiaries, but also for the market to grow.
“You have to start with the use case,” Jain points out. “If you tell me you have a project that starts at age two, that’s a solved problem, solutions are available.”
Still, the availability of solutions is not in and of itself enough, Jain concedes. “It all boils down often to the human factor. The interaction of the finger with the fingerprint reader, and the operator training.”
Finger placement is a persistent issue for adult enrollment and matching, according to Grigore. Data analysis from Simprints’ early pilots showed that incorrect and inconsistent finger placement was the main cause behind a lower than expected match rate. The company designed its device with a ridge at the end of the scanner surface so that people can be told to place their fingers all the way to the end, and also signals users with a colored light when the finger is in place.
The requirements, Jain notes, are very different from those of the scanners used by law enforcement agencies in controlled environments, such as were used to build the second largest biometric database, held by the FBI. They are also very different from the sensors for smartphone integration that are the product-development focus of many biometric technology providers.
Both Grigore and Aronoff-Spencer emphasize the role of design elements in optimizing the quality of the prints acquired.
In early trials, Simprints found that multiple commercially available scanners broke down within days of being introduced to developing world conditions. The manufacturers claimed the devices were not meant for outdoor use, Grigore says, but that fact was not mentioned on their accompanying data sheets. A significantly new kind of device was necessary.
“We’ve done a lot of field research, and we employed a lot of current design methodologies which are not very often employed in the biometrics world, from what I’ve seen,” Grigore says.
Those methodologies included taking data from focus groups, one on one interviews, shadowing, intuitive exercises, and co-creation exercises. Common design methods, including human centered design concepts, were applied across multiple stakeholder groups to come up with design criteria such as using two indications to mean the same thing. A color plus a symbol may indicate capture success, for example.
Kidprint realized early that the interaction with a flat pane of glass was impeding the capture of quality fingerprint data from newborns. From that point of departure, design efforts included “little rubber donuts,” fins to guide finger placement, and a variety of different apertures. The group also experimented with different types of internal refraction, different polarizations, and different kinds of light to improve data capture. “The technology was the problem,” Aronoff-Spencer explains.
If the design and training are effective enough, the data quality will enable a high enough match rate to make the business case.
Simprints’ hardware design went through roughly ten iterations, and its software was similarly overhauled in development to address the user interface and experience aspects of capturing good fingerprint quality. “It’s easy if you do it right,” Grigore says. “With biometrics, it’s so easy to do it wrong. It’s so easy to capture bad quality, whether its face, iris or fingerprints.”
Fingerprints are the only option on the horizon for biometric birth registries or to biometrically identify very young patients for vaccination programs. If capturing good quality biometric data from adults in developing world contexts is difficult, are fingerprints under two years old really feasible?
For infant fingerprints, Jain and Aronoff-Spencer both say that it is possible in theory to enroll the fingerprints of newborn babies, and may be possible in practice in the near future, but they differ on the importance of higher sensor resolution. Jain has been working with sensors up to 1900ppi designed by his team for a trial with infants at a hospital in Agra, India. Previously, the team used an NEC sensor with 1200ppi in research which indicated that at least 1000ppi is necessary to capture high-quality fingerprints from infants.
“My sense is there’s gonna be a next generation of contact scanners coming out,” Aronoff-Spencer says. “Like Jain’s work with NEC, you’ll start to see things more like 1260ppi et cetera, and I honestly think those will work well for kids a year and older, and they work ok for kids six months and older, but they start to fall apart under six months because that’s really the age where kids’ fingerprints are mushy.”
The Kidprint team has found that infants have a thin sheen of skin which tends to block the fingerprint, but peels within a few days of birth. This causes an inflection point in the performance, but even before that, the failure rate within the first three days is below 10 percent with the false acceptance rate (FAR) set to 10-4, with the failures skewed towards the first 12 hours after birth, according to Aronoff-Spencer.
“Eventually other techniques that allow subsurface imaging will improve on that, but for now we think our tech is working with a pretty low failure-to rate and a pretty high match rate right down to birth,” he says.
For his part, Jain wants to see published data on the error rates Kidprint is achieving. Aronoff-Spencer says his group is working towards publishing his data. Jain stresses the importance of longitudinal data, without which the practical effectiveness of fingerprinting technologies or devices is speculative. Jain’s team is collecting data from infants at hospital visits in Agra, India, spaced three months apart for the first year of their lives.
Actually delivering the systems to support developing world biometric identification programs also means meeting stringent cost requirements. Ion was created as a B-Corp to sell its device as a bespoke solution for various markets, with the goal of driving down cost for developing world projects through scale, Kleeman explains. Ion is discussing target costs, and its licensing plan is intended to help keep manufacturing costs low. The device is also different from most biometric hardware in that it is multi-functional, confirming the patient’s identity and then taking for instance temperature, heart rate, pulse, oxygen and hemoglobin levels.
“That allowed us to create a different solution set which has very different economics and impact,” Kleeman says.
Jain’s team has developed an ergonomic scanner with an open-source design, which can be constructed from parts bought on Amazon for around $85, he says, and can be put together in an hour. In the right circumstances, Jain says it could be produced for as little as $30.
Simprints uses a different model, with price determined on a per-project basis, determined in consultation with the government ministry or institutional client. Budgets are typically set for a five-year period, and cover training and support, along with the hardware. Manhart says that once the agreement is in place, Simprints will do whatever it takes to achieve the project goals, including, if necessary, taking a loss.
Another aspect of Aadhaar’s success that Jain identifies, and one which has potentially major implications for biometrics vendors, is the way the system was developed. By setting up a dedicated agency (the UIDAI), and appointing tech industry leader Nandan Nilekani as its Chairman, the Indian government empowered the agency to build a network of experts to do the work internally.
“They designed the system,” Jain points out. “They did not rely on the vendors to design the system, or a private consulting company.”
When service providers were brought in for the roll-out, three different vendors were given equal share and paid based on the number of transactions. The share of each was then adjusted based on performance, providing further incentive to keep improving their algorithms. The pay of the registrars who the enrollment was contracted out to was also adjusted according to data quality.
“When you talk about innovation in biometrics, it’s not always innovation in terms of resolution and so on,” Jain stresses.
At a small scale, funders will back projects based on potential, but Manhart reports that governments are now prepared to make their own decisions about biometric technology choices. That means as a biometrics provider, it is the government, rather than funders, who have to be convinced that a certain technology is the right one.
“The days are over in most places where a Western funder decides to throw a hundred million at something and the government has to accept it,” he asserts.
Governments, international development and humanitarian organizations may be ready to meet the challenges of providing biometric identification in the developing world, but industry engagement is uneven, with most new research and development dedicated to addressing established, profitable markets. There are a few other companies working to bring innovation to the space. Element is addressing the challenges of remote access by leveraging the ubiquity of the cellphone camera, and has partnered with the Bill Gates-backed Global Good Fund to work on infant biometrics. Everest ID combines biometrics with a compartmentalized blockchain system for its globally inclusive identity system. Some other companies have developed innovative approaches to dealing with developing world challenges.
Unfortunately, when Jain reached out to many companies through his extensive connections six years ago, he found that unless an agency was dangling a large amount of money, there was little interest in the developing world. “They all sort of indicated unwillingness, that is there is not enough business for them,” he says.
With UN sustainable development goal (SDG) 16.9 calling for universal legal identity including birth registration by 2030, the motivation is clearer than the market. Some 736 million people live on less than $1.90 a day, according to recent World Bank statistics, but as governments and international agencies will likely pay the majority of the bill for achieving the SDG target, it is they and the solution providers who must find a workable value proposition. If the development community is right, then project success will help spark a virtuous cycle in which the buying power of people who were previously invisible to governments and financial institutions will increase rapidly. If this happens, then it is likely that a significant amount of market share will be claimed early in the growth period, leaving those companies that wait at risk of missing out entirely.
“Our industry is building very powerful tools that we fundamentally believe can have a huge positive impact in many areas, whether its healthcare, education, or humanitarian relief for disasters,” Manhart contends. “We would encourage an enhanced focus by the industry on this potential, and we join as many organizations as possible to join our mission. We have a huge problem to solve and we can’t do it alone.”
Anil Jain | biometrics | digital identity | fingerprint technology | Gates Foundation | healthcare | Identification for Development (ID4D) | infant biometrics | Ion Biometrics | SimPrints | UIDAI | World Bank
One Reply to “How many billions are the last billion’s biometrics worth?”
Great post! Very informative.