The recent explosion in cheap computing power, development in algorithms, and the increase of available data are making new classroom learning tools possible. Photo credit: Getty images

AI Technology is Disrupting the Traditional Classroom. Here’s a Progress Report.

<p>Technology has fundamentally transformed most aspects of modern life, from how we travel to our healthcare. But there’s one place that remains markedly unchanged: the classroom.</p><p>The schoolrooms of today are strikingly similar to those of 50 years ago. Students sit in a room together and complete the same lessons—typically using the same textbooks—no matter their learning styles or mastery of the topic. Some students get left behind. Others are left unchallenged and bored by this middle-of-the-road approach.</p><p>AI could change that. The field of AI in education, or AIEd, started around 30 years ago, with systems designed to teach basic skills, like arithmetic, while recognizing and addressing common mistakes. More recently, the explosion in cheap computing power, development in algorithms, and the increase of available data are making new, more sophisticated tools possible.</p>
<div class="related"></div>
<p>“You’ve got a perfect storm, really,” says Rose Luckin, a professor at University College London who has studied AIEd for the past 20 years. “You can do things that you weren’t able to do before.”</p><p>AIEd now helps investigate the steps students go through when learning subjects from calculus to chemistry, shining a light on what individual learners need to progress. To get there<b>,</b> an AI program is first trained on hundreds or thousands of students’ work, gaining a knowledge base of the common areas that give learners trouble. Then over time, as an individual uses the system, the AI homes in on specifics to focus on, usually offering bespoke lessons to brush up on skills, and, in some cases, offer pep talks through bots. These insights are also given to teachers for more focused help.</p><p>Math is a particularly ripe subject matter for AI to tackle because it follows sets of rules that are easier for AI to understand—compared to something like creative writing, which has more subjective criteria for success. Math is also an area in which American students are falling behind their international peers. <a href="http://www.pewresearch.org/fact-tank/2017/02/15/u-s-students-internationally-math-science/">According to one survey</a>, American 15-year-olds are 38th out of 71 in the world for mathematics among developed countries.</p><p>Companies like Thinkster Math and Carnegie Learning are designing tools to help students learn math, as well as provide more insights for instructors. Every response by a student updates Thinkster Math’s personalized algorithm to create a “brain fingerprint,” which helps recommend the next set of assignments, whether it be remedial practice or moving students along. Teachers have the option to reject the suggestions and provide a different set of work for students.</p><p>“We can’t rely on AI alone,” says Raj Valli, founder and CEO of Thinkster Math.</p><p>AI-powered assistance is also being applied to language acquisition. For example, Duolingo is a free app that offers curriculum for learning 33 languages for English speakers. It includes a Duolingo for Schools program, which gives students personalized lessons and feedback in over 300,000 classrooms around the globe.Since Duolingo’s launch in 2011, it has incorporated more machine learning into every segment of its work.</p><p>Though not as good as moving to a new land and being immersed in a language, Duolingo is accessible and free, and has gathered enough data to know when to present new concepts to the user, quickly figuring out what users already know.</p><p>“After five minutes, we could put you halfway through the course if that is where you belong,” says Burr Settles, the research director at Duolingo.</p><p>The app also recently introduced an <a href="http://bots.duolingo.com/">AI bot</a>, which lets users practice speaking a new language. Duolingo found that when people talk to a chatbot, they don't feel as embarrassed as when practicing with a human.</p>
<figure class="richtext-image full-width">
<img alt="" class="richtext-image full-width" src="https://www.pbs.org/wgbh/nova/media/images/Duolingo1.width-990_XPnTn79.jpg"/>
<figcaption><p>Duolingo, a free app that offers curriculum for learning 33 languages for English speakers, has gathered enough data to know when to present new concepts to the user. Image credit: Emily Zendt, WGBH</p>
<div class="credit"><p></p></div>
</figcaption>
</figure>
<p>That kind of one-on-one tutoring has a long history of giving students an advantage (see: Plato, Aristotle, Socrates). The problem is, individualized tutoring doesn’t scale if only human teachers are available—in many cases, there simply aren’t enough of them to go around. AI could help fill the gap by simulating one-to-one human tutoring, delivering activities paced to a learner’s abilities.</p><p>AIEd doesn’t mean cutting teachers out: Instead, they would be freed up to work on social skills. After all, most researchers working on different technologies say that AI isn’t anywhere close to good enough to do all the things we rely on human teachers for.</p><p>“There are some things that the computer is really good about, like collecting data points, and there are some things that teachers are really good at, like motivating students,” says Steve Ritter, the co-founder of Carnegie Learning.</p><p>Carnegie Learning is also working with <a href="http://bit.ly/2G6zEsM">Lumilo</a>, a startup building an AI and augmented reality (AR) assistant that will keep teachers in the loop as students work on assignments. Built with input from teachers and tested in more than 30 classrooms around Pittsburgh, Lumilo allows teachers to monitor students' progress in real time. Information on how students are doing is picked up by the AI and projected above each of their heads, visible only to the teacher wearing AR-powered glasses. The teacher is able to pinpoint who may need a more challenging assignment and who might be struggling, even if those students aren’t asking questions.</p><p>“Without real-time analytics, the teachers don’t have a lot to go on," says Ken Holstein, a doctoral candidate at CMU and one of the creators of Lumilo. “Their attention goes to the hand raisers.”</p>
<div class="ad"></div>
<h2>Pitfalls of Technology in Classroom Settings</h2><p>Despite being worked on for decades, AI hasn’t made huge inroads into schools yet. That is partially due to how conservative the field is compared to other areas that are being upended by AI, Ritter says. Limited resources are also a factor, as school districts with tight budgets are naturally reluctant to invest in risky new technologies, however promising they might be.</p><p>“We aren’t competing against other software,” Ritter says. “We are competing against textbooks.”</p><p>But AI in classrooms is also slowed by concerns over equity, one of the biggest problems facing the field of education. Intelligent tutors could be used to help close the achievement gap, or it could replace teachers for poor students while wealthy students get a more balanced mix of human and AI help. Where some see the potential for AI—like other new technologies—to make education more equitable, others fear it will do the opposite, further widening disparities between rich and poor schools.</p>
<figure class="richtext-image full-width">
<img alt="" class="richtext-image full-width" src="https://www.pbs.org/wgbh/nova/media/images/ipad-1126136_960_720.width-990_ztpq7gU.jpg"/>
<figcaption><p>One of the biggest problems facing the field of education is equity. AI could transform traditional classrooms on a global scale—as long as its technology remains accessible. Image credit: SchoolPRPro, Pixabay</p>
<div class="credit"><p></p></div>
</figcaption>
</figure>
<p>AI-powered tools also run the risk of creating classroom-based, mini-surveillance states. <a href="https://www.scmp.com/news/china/society/article/2146387/pay-attention-back-chinese-school-installs-facial-recognition">At least one Chinese high school</a> is currently using facial-recognition technology that scans students every 30 seconds, looking to see if students are paying attention or losing interest. The privacy of a student’s data is also a huge concern. A hack could expose sensitive information or be used without permission later in a person’s life: Imagine your high school physics test result being something that potential employers could look up.</p><p>Bruce McLaren, the president of the International Society for Artificial Intelligence in Education and a research professor at Carnegie Mellon University, says that after speaking at a recent conference about AIEd, almost every question was about the intrusion of tech in classrooms. He says it’s a valid concern, but believes AI is a net positive and probably inevitable anyway.</p><p>“It’s the way the world is going,” McLaren says. “And education will not be the exception to that.”</p><h2>Learning AI to Avoid Being Replaced by AI</h2><p>Besides AI changing how students learn, schools are gearing up to teach students <i>about</i> AI. The Montour School District outside of Pittsburgh is the first in the country to create an AI curriculum in a public school. Justin Aglio, the Director of Academic Achievement and District Innovation of Montour School District, says the curriculum launched just a few weeks ago and that he and his colleagues are already getting calls from around the country asking how to roll out similar programs.</p><p>“AI is not part of the future anymore,” he says. “It’s the present.”</p><p>The inaugural first class, which is starting in one middle school but will be rolled out to the rest of the 2,900 students in the district if things go well, is on ethics in AI. Other plans include an autonomous robotics class with <a href="https://www.anki.com/en-us/cozmo">Cozmo</a>, a small, programmable robot, and a music project using AI-powered <a href="https://www.ampermusic.com/">Amper</a>. Algio says he and his colleagues are also planning field trips to local companies working in AI.</p><p>Most importantly, the AI lessons will be integrated into classes that all students take so it’s not just a self-selecting, privileged group willing to try out a new curriculum. “I’ve seen too many times where you create these fancy classes that intimidate students and you have the same population taking the class,” he says.</p><p>The first students going through the program are graduating from high school in the early 2020s. By then, Aglio says, the world will be totally changed by AI, and it’s education’s job to make sure students are ready.</p><p>“If we aren’t teaching it to them now,” Aglio explains, “they are going to be dinosaurs by the time they get out.”</p>
<div class="newsletter"></div>

<p>Artificial intelligence has the potential to personalize learning at scale. The challenge: making sure it benefits everyone.</p>

AI technology is being developed to help rangers detect poachers, who represent a serious threat to African elephants, like this mother and calf in Ol Pejeta Conservancy, Kenya. Image credit: Allison Mitchell.

Can Artificial Intelligence Help Save the Natural World?

<p>African elephants, the biggest land animals on earth, are in grave peril: They could be extinct in <a href="https://www.express.co.uk/news/world/421411/Elephants-extinct-within-12-years">the next decade</a>. Poachers kill an <a href="https://www.theguardian.com/environment/2017/oct/24/elephant-poaching-drops-africa-but-populations-continue-to-fall">estimated 55 elephants</a> every day.</p><p>The challenge of protecting elephants is vast, literally—reserves are sprawling, often remote, and understaffed. AI is starting to fill the gaps by being the eyes and brains behind spotting or predicting where the poachers might be. One of the most recent projects to stop attacks is <a href="http://www.resolv.org/site-BiodiversityWildlifeSolutions/trailguard/">TrailGuard AI</a>, a system of small cameras with algorithms for image detection and object recognition built in.</p><p>The name TrailGuard comes from the fact that despite the enormous areas some reserves cover, there are still choke points poachers have to travel through, allowing cameras to be setup up strategically in those places. To be helpful, the cameras have to work autonomously, be low-powered, and send photos in real time.</p>
<div class="related"></div>
<p>An earlier iteration of the TrailGuard camera didn’t have AI built in and sent back photos that 75 percent of the time had no humans in them. The cameras would be triggered by a cloud moving in front of the sun, swaying grass, or animals going by. These false triggers eat up precious battery life and become a nuisance to the rangers who have to drop what they are doing and look through images. Plus, changing batteries every few weeks puts people in danger and potentially gives away where the cameras are set up.</p><p>Adding AI that detects whether a person is in the photo reduces the error rate to a fraction of previous attempts—and it improves its accuracy with time. Capturing and sending fewer images means TrailGuard cameras now last up to 18 months without having to swap the batteries out.</p><p>The AI behind the smart cameras has been used in urban settings for years but hasn’t made an appearance to stop poaching until now. “It’s never been used in national parks where we need it the most," says Eric Dinerstein, the Director of Biodiversity and Wildlife at RESOLVE, a conservation nonprofit and a partner on the TrailGuard AI project.</p><p>For cash-strapped national parks trying to stop well-funded poaching forces that can include anything from machine guns to <a href="https://www.africanparks.org/newsroom/press-releases/update-garamba-national-parks-poaching-crisis">helicopters</a>, any help is appreciated.</p><p>“This is a huge advantage for conservation,” says Alex Dehgan, the CEO of Conservation X Labs, a nonprofit looking to end human-driven extinctions with technology.</p><p>Dehgan, who was not involved in the TrailGuard project, points out that rangers will still be crucial to the conservation efforts, but will now have more information to act on than ever before.</p><p>“[It’s] going to make us all into the equivalent of superheroes,” he says.</p><p>The goal is to deploy TrailGuard AI in 100 reserves in Africa by the end of the year, starting in the national parks in the Serengeti and Garamba. Ultimately, there are plans to take the technology to Southeast Asia and South America—and possibly use it to combat issues other than poaching, like illegal logging.</p><p>“We think it's going to be a game changer,” Dinerstein says.</p>
<figure class="richtext-image full-width">
<img alt="" class="richtext-image full-width" src="https://www.pbs.org/wgbh/nova/media/images/Intel-anti-poaching-4.width-990_3kXNHw2.jpg"/>
<figcaption><p>TrailGuard AI, developed by Intel and conservation nonprofit RESOLVE, is a system of small cameras that can help park rangers identify possible poachers quickly, without human image analysis. Image credit: RESOLVE</p>
<div class="credit"><p></p></div>
</figcaption>
</figure>
<h2>Saving the Natural World with Artificial Intelligence</h2><p>Humanity is making it tough for the rest of the inhabitants on Earth to thrive. Deforestation, poaching, and urban sprawl are just some of the problems we’ve unleashed on our non-<i>homo sapiens</i> neighbors. Those pressures, as well as others like climate change and pollution (which aren’t good for humans either), are contributing to a loss of plant and animal species that scientists argue is nothing less than a massive and human-made <a href="https://en.wikipedia.org/wiki/Holocene_extinction">sixth extinction</a> event.</p><p>Conservationists have long turned to technology—such as remote sensors and animal I.D. tags—to learn more about the species they study, bring attention to the worst problems, and inform solutions like protected refuges that can help species return to health. AI and machine learning are the latest technologies at their disposal. And luckily for perennially strapped conservation efforts, better-funded AI research is creating free tools, cheap cloud services, and open source resources that put cutting edge technology within reach for even small nonprofits.</p><p>The need for large datasets, however, still holds backs many ecological AI projects. Most of the data that does exist for conservation isn’t digitized, is incomplete, or is housed in proprietary databases locked down at universities.</p><p>Projects like <a href="https://www.wildme.org/">Wild Me</a> are trying to address these problems by aggregating large amounts of data for scientists. Wild Me lets researchers upload photos of the animals they study to an open-source platform, and scrapes large amounts of data from websites like YouTube and Flickr. When safari-goers upload footage of a pack of zebras they saw, for example, Wild Me’s computer vision algorithm spots the zebras and can identify individuals it has seen before, creating a record that lets biologists study the health and habitat of animals in new ways.</p><p>Another data creation effort is <a href="http://conservify.org/">Conservify</a>, a lab dedicated to using technology to democratize data gathering to make an impact on conservation efforts. One way they’re doing that is by building low-cost sensors that can be placed in the wild to gather new data. They are also developing an online platform called FieldKit that will help researchers, students, and weekend naturalists share data.</p><p>Trying out new technology does come with hazards. Communities aren’t going to be happy with tech that doesn’t work or affects them negatively. It might also disturb wildlife unaccustomed to its presence.</p><p>“Once you go put it out in the field, that’s where the real risk happens,” says Shah Selbe, Conservify’s CEO and founder.</p><p>Selbe points to the explosion of poorly-trained drone operators as an example of technology sometimes being put to use before it’s ready. Drones have major potential for conservation research, especially when combined with computer vision algorithms that can swiftly parse hours of footage, but drones need to be used carefully so as not to bother local communities or animals. Fortunately, Selbe says so far he hasn’t seen significant failures with new AI tools. The stakes are high because AI, done right, could open the world up to further exploration and understanding.</p><p>“In terms of the work that I do, everything has absolutely changed,” he says.</p>
<div class="ad"></div>
<h2>Discovering Endangered Plants Before it’s Too Late</h2><p>In a fluorescent-lit room lined with gray filing cabinets at the University of Maryland (UMD), a hundred thousand plant specimens are tidily housed in manila folders. The samples in the Norton-Brown Herbarium have been carefully collected for more than a century from across the globe, although the majority are from the mid-Atlantic region. One such sample is <i>Phacelia covillei,</i> a flowering plant with bottle-shaped blue blossoms found in scattered pockets in states like Maryland, North Carolina, Indiana, and Missouri.</p><p>Due to factors like urban development and climate change, <i>Phacelia covillei</i> is at risk of extinction, according to <a href="http://bit.ly/2T9ZKSC">NatureServe</a>, a ranking system used to denote the relative imperilment of different plant and animal species in North America. It made the list thanks to a new machine learning tool that tries to identify threatened plants before it’s too late to save them.</p>
<figure class="richtext-image full-width">
<img alt="" class="richtext-image full-width" src="https://www.pbs.org/wgbh/nova/media/images/IMG_3634.width-990_WsyObbV.jpg"/>
<figcaption><p>A new machine learning tool aims to speed up the identification of threatened plants like <i>P. covillei</i>, pictured here from the University of Maryland's Norton-Brown's Herbarium. Image credit: Jackie Snow</p>
<div class="credit"><p></p></div>
</figcaption>
</figure>
<p>It’s a big challenge. The vast majority of plants haven’t been assessed at all: Less than 10 percent of plant species around the world have been checked for the IUCN Red List, the premier directory for global extinction risks that is often used to make conservation decisions. Anahí Espíndola, a professor of evolutionary ecology at UMD who worked on the machine learning tool, wanted to see if she could find ways to zero in on those most in need of classification.</p><p>Espíndola and her team trained a machine learning algorithm on details known about plants that are already on the IUCN Red List—like their location, range, and physical traits—and used it to assess 150,000 plant species whose vulnerability is currently unknown. Their method found that more than 10 percent of those unassessed plants were at risk.</p><p>This AI tool could speed up conservation efforts and direct limited resources more effectively than current practices. Geographic bias is one problem: Plants are better studied in the U.S. and Europe because of bigger government and university research budgets. Espíndola says AI can quickly pinpoint the plants most at risk, regardless of where they might be. AI might also prevent another problem that plagues animal conservation: The cutest species get an outsized amount of the attention.</p><p>“You don’t get that bias [with AI],” she says of our preference for the cute and cuddly. “It tells you where to go first.”</p><h2>Seeing the Forest for the Trees</h2><p>Conservation AI isn’t just for remote areas; it could also make cities greener. For example, trees reduce the <a href="https://www.epa.gov/heat-islands/using-trees-and-vegetation-reduce-heat-islands">urban heat island effect</a>, <a href="https://www.dec.ny.gov/lands/90720.html">reduce stress</a> in humans, and <a href="https://www.bbc.com/news/science-environment-35777927">reduce flooding during storms</a>. Despite all of these benefits, U.S. cities are losing 36 million trees a year, according to a <a href="https://www.sciencedirect.com/science/article/pii/S1618866717307094?via%3Dihub">study from the Forest Service</a>.</p><p>Knowing where the trees are (or aren’t) isn’t as easy as it sounds. Counting from the ground takes years and the results are often incomplete due to difficulties distinguishing trees on private property, or using satellite images that aren’t crisp or detailed enough for the human eye. Geospatial analytics startup Descartes Labs built a machine learning tool that can take one-meter resolution satellite imagery and count individual trees, without getting tricked by non-tree greenery or shadows that might appear to be trees. This would point out areas that are “tree deserts” and, over time, show changing areas that might need renewed attention.</p><p>Descartes Labs originally developed the tool to <a href="https://www.bloomberg.com/news/articles/2018-04-11/what-satellites-can-t-see-u-s-crop-forecasters-walk-and-learn">predict corn harvests</a> but could see the system being used as a deforestation prevention tool. Eventually, Descartes Labs might create layers of different informational maps that would be useful to people trying to manage land for all sorts of reasons, including conservation efforts.</p><p>“You can be surprised by what you could see by combining layers,” says Tim Wallace, a geographer at Descartes Labs. “The more we combine, the more powerful it becomes.”</p><h2>Era of AI-Powered Conservation</h2><p>It’s not just ecologists and startups working on saving the world. Some of the biggest companies in the world like <a href="https://www.microsoft.com/en-us/ai-for-earth/apis-and-applications">Microsoft</a>, <a href="https://www.intel.ai/ai4socialgood/#gs.rl2xbVp2">Intel</a>, and <a href="https://ai.google/social-good">Google</a> all have AI for good projects, including some that target conservation issues. Not only is this work potentially good for the planet, but it’s good for business, too. Creating cutting-edge tools and putting money into conservation programs is a great way to recruit top talent that doesn’t want to just work on improving advertising click-through rates. And there is an additional bonus: If technology can work in the field, it should also work in less rugged office and factory settings.</p><p>“[Conservationists] always set the bar super high,” says Lucas Joppa, Microsoft's Chief Environmental Scientist. “And if you can jump that bar, you passed the bar for many, many, many other use cases by far.”</p><p>Joppa has experience with clearing those bars with Microsoft’s <a href="http://bit.ly/2Er5b7M">AI for Earth</a>, a five-year, $50 million initiative to back projects by environmental groups and researchers working on sustainable plans using AI. Joppa, who helps oversee the program, says that there is nothing harder or more important than creating technology that can solve these environmental problems.</p><p>"People think that the hardest problem is how to route an autonomous car to the closest latte,” he says. “We have other problems we need to tackle.”</p><p>Of course, there is always the concern that AI could do more harm than good for the environment. Previous industrial revolutions were driven by new technologies that ended up being harmful to the planet, from factories spewing smog to encouraging mining of <a href="https://news.vice.com/en_us/article/qv5pxm/theres-gold-in-those-cell-phones-but-it-comes-at-a-steep-price-to-the-environment">precious metals for smartphones</a>. This AI revolution, Joppa says, will have to be carefully designed to benefit the planet and not create problems, unanticipated or otherwise, for future generations.</p><p>“That’s not easy to do,” he says.</p><p>To learn more about researchers' testing of TrailGuard AI, which could someday help rangers better protect elephants and other endangered species, watch this video:</p>
<div class="cove-video">
<iframe allowfullscreen="" class="inline-cove-player" frameborder="0" height="100%" id="partnerPlayer" marginheight="0" marginwidth="0" scrolling="no" src="//player.pbs.org/partnerplayer/caP1Y7S0kt2iv40hAZ0C8g==/?topbar=false&amp;end=0&amp;endscreen=true&amp;start=0&amp;autoplay=false" width="100%">
</iframe>
</div>
<div class="newsletter"></div>

<p>AI-powered tools are giving conservationists new ways to combat the daunting, ongoing, human-caused problem of mass extinction. It won’t be easy.</p>

With an effort called Community Control Over Police Surveillance (CCOPS), the ACLU is attempting to give citizens more control over the technology that police could use to monitor them. Image Credit: iStock, asxsoonxas

Communities come face-to-face with the growing power of facial recognition technology

<p>When Keith Rose joined the Occupy St. Louis movement in 2011, he didn't heed the advice of a fellow demonstrator to wear bandanas to cover his face. The police had cameras out, but Rose didn't think they were doing much with the photos they were taking.</p><p>When he joined the Ferguson protests a few years later, it seemed to him that law enforcement had indeed started to put together a file on him. He had not been arrested at that point, but after a few run-ins with police where they knew things like where he had been, who he had been with, or what he had said, he thought it seemed pretty clear that they were monitoring him.</p>
<div class="related"></div>
<p>"It really upset me and got me more concerned with surveillance," Rose says.</p><p>During the protests, the St. Louis Metropolitan Police Department opened a surveillance hub called the Real Time Crime Center (RTCC). This data center aggregates information gathered through technologies like license plate readers, sensors that can detect and locate gunfire, and cameras spread across the city. Most of RTCC, as well as much of the technology that powers it, was not funded by the city, but through grants and public-private partnerships with companies like Motorola. That approach, according to the American Civil Liberties Union (ACLU), allowed the police department to <a href="https://www.aclu.org/blog/privacy-technology/surveillance-technologies/how-shift-power-police-people">sidestep going to the city's elected Board of Aldermen</a> for approval and city funds.</p><p>"It's really disturbing what this says about the future of big tech being able to expand its own power," Rose says.</p><p>But with an effort called <a href="https://www.aclu.org/issues/privacy-technology/surveillance-technologies/community-control-over-police-surveillance?redirect=feature/community-control-over-police-surveillance">Community Control Over Police Surveillance</a> (CCOPS), the ACLU is attempting to give citizens more control over the technology that police could use to monitor them. Launched in 2016, CCOPS aims to help local communities draft and pass laws giving them oversight of any surveillance technology that law enforcement agencies or the government want to deploy.</p><p>The ACLU works with local lawmakers to craft each law, changing it to meet the community's specific needs. A dozen cities, plus San Francisco's BART system, have passed CCOPS laws. In addition to St. Louis, at least another dozen cities, plus the state of Maine, are in the process of adopting their own versions of the law.</p><p>St. Louis Alderman John Collins-Muhammad helped <a href="https://www.stlouis-mo.gov/government/city-laws/board-bills/boardbill.cfm?bbDetail=true&amp;BBId=10831&amp;fbclid=IwAR2dISUYbLnHGl-n7Tys97ijbIKg8rbMMWR9xdxZZI24kjj1o0bZZoFWTPU">introduce CCOPS legislation</a> in 2017 and 2018, and recently <a href="https://www.stlouis-mo.gov/government/city-laws/board-bills/boardbill.cfm?bbDetail=true&amp;BBId=13371">reintroduced</a> an updated version. Collins-Muhammad said that he and the rest of the Board got community input and updated each iteration. He thinks the latest version of the bill will pass this year. The law would create a process for board approval of any new surveillance technology, require annual reports for equipment purchases, strengthen sunshine laws around law enforcement data, and call for publishing crime reduction statistics. The bill tries to balance the concerns of both the public and the police, Collins-Muhammad says.</p><p>"We don't want to invade privacy," he says. "But we also don't want to limit the police in doing their jobs."</p><p>The St. Louis Metropolitan Police Department has also contributed comments on the bill and has been supportive of it in general, Collins-Muhammad says. In a statement, Keith Barrett, a sergeant with ‎the Department, said that civil liberties and surveillance technology can coexist if used in a responsible manner while still helping police do their job.</p><p>“The advancement of technological resources continues to allow our department to deter, document, and reduce crime,” Barrett says. “The technology that is deployed continues to aid our efforts to prevent crime and apprehend criminals. We look forward to working with all our community partners in ensuring transparency and accountability.”</p><p>Privacy Watch, a group of nonprofits in St. Louis, took part in the discussion around the bill. It wanted the law to address issues related to bias with the technology, privacy, freedom of speech, and the cost to taxpayers.</p><p>"There are so many layers of concern," says Kendra Tatum, a St. Louis resident and an organizer for the Organization for Black Struggle, a member group of Privacy Watch. "People refer to the phrase ‘Big Brother’ jokingly a lot, but we all know that this type of stuff happens."</p><p>Collins-Muhammad says he knows the threat personally, having experienced surveillance when he was an activist during the Ferguson protests. He said it was “scary” and got him interested in creating policy that could limit that experience for others in the future.</p><p>"This is something very important to me," he says.</p><h3><b>An AI Panopticon</b></h3><p>While the Real Time Crime Center doesn't use AI-powered tools yet, it is technically easy to overlay face recognition software after cameras are in use. The proposed CCOPS legislation in St. Louis wouldn't outlaw the technology, but would require law enforcement to go through the Board of Aldermen if they want to use it.</p><p>AI-powered surveillance technology has <a href="https://carnegieendowment.org/publications/interactive/ai-surveillance">spread across the world quickly</a>. ACLU'S <a href="https://www.aclu.org/report/dawn-robot-surveillance">Dawn of Robot Surveillance</a> report points out that this is occurring as camera technology is improving with higher resolution and night-vision capabilities. At the same time, more cities are installing centralized surveillance systems, with about <a href="https://www.sdmmag.com/articles/92407-rise-of-surveillance-camera-installed-base-slows">350 million surveillance cameras</a> across the world in 2016.</p>
<figure class="richtext-image full-width">
<img alt="" class="richtext-image full-width" src="https://www.pbs.org/wgbh/nova/media/images/surveillance2.width-990_kwU3fnN.jpg"/>
<figcaption><p>With artificial intelligence, a lot of surveillance work can be automated, making AI game-changing technology. Image Credit: iStock, andreusK</p>
</figcaption>
</figure>
<p>The potential of facial recognition technology is perhaps most prominently on display in China. The government not only uses it to give out traffic tickets, and has passed laws mandating face scans for those signing up for internet or a new phone number, but has put it to work <a href="https://www.nytimes.com/2019/04/14/technology/china-surveillance-artificial-intelligence-racial-profiling.html">monitoring and imprisoning the Uighur minority</a>. By 2020, China plans to have more than <a href="https://www.cbsnews.com/news/china-social-credit-system-surveillance-cameras/">600 million cameras operating</a>.</p><p>Beyond concerns over increased surveillance, facial recognition software has been shown to perform less well on dark skin. In a 2018 report called "<a href="http://gendershades.org/">Gender Shades</a>," researchers Joy Buolamwini and Timnit Gebru found that three commercially available systems performed no better than chance at identifying women with darker skin tones.</p><p>The ACLU did its own study showing that Rekognition, Amazon's face recognition software, <a href="https://www.aclu.org/blog/privacy-technology/surveillance-technologies/amazons-face-recognition-falsely-matched-28">misidentified 28 members of Congress</a>. Forty percent of the algorithm’s mistakes were made in attempting to identify people of color, even though they make up only 20 percent of Congress. Amazon disputed the study and showed that with the recommended higher confidence threshold, which only lets the algorithm offer up matches if it is very sure it is right, the misidentifications disappear. However, one of the police departments<a href="https://gizmodo.com/defense-of-amazons-face-recognition-tool-undermined-by-1832238149">testing the software admitted</a> to not setting any confidence threshold.</p><p>That isn't the only case of police departments using software in a way that the developers didn't intend. A study by Georgetown Law Center on Privacy and Technology <a href="https://www.flawedfacedata.com/">found that police fed celebrity photos</a> into face recognition software, along with sketches and photoshopped images.</p><p>There is a long history of surveillance in the U.S. according to Matt Cagle, an attorney at the ACLU of Northern California. Public and civil rights groups like the ACLU have fought against much of it, with varying levels of success. With AI, however, a lot of surveillance work can be automated, making AI game-changing technology.</p><p>"If they can watch us, they can control us," Cagle says.</p><p>In fact, the face recognition conversation can be part of the larger conversation about government surveillance according to Chad Marlow, a senior advocacy and policy counsel at the ACLU, who is also overseeing CCOPS. The proliferation of technology like surveillance cameras and Stingrays—which mimic cell phone towers and send out signals that trick nearby phones into transmitting identifying information and data about their locations—can be reevaluated. Also, anything new developed, with or without AI capabilities, can be integrated in a way that the community is aware of and has some control over.</p><p>“In that way, we aren’t playing whack-a-mole,” Marlow says.</p>
<div class="ad"></div>
<h3><b>Drawbacks of Bans</b></h3><p>Four cities have gone further than community oversight with their CCOPS laws: Berkeley, Oakland, San Francisco, and Somerville, Massachusetts have outright banned face recognition by the government. Earlier this year, the ACLU, along with another 60 groups, <a href="https://www.aclu.org/letter/coalition-letter-calling-federal-moratorium-face-recognition">sent a letter to Congress</a> asking for a federal moratorium on the use of facial recognition for law enforcement and immigration enforcement until Congress fully debates which practices should be allowed.</p><p>“This capability threatens to create a world where people are watched and identified as they attend a protest, congregate outside a place of worship, visit a medical provider, or simply go about their daily lives,” the letter said.</p><p>But Daniel Castro, vice president of the Information Technology and Innovation Foundation and director of the Center for Data Innovation, says cities could also end up blocking better potential uses of face recognition tools out of fear of a surveillance state.</p><p>"We're not there yet," he says. "A lot of the concerns there are premature."</p><p>Other law enforcement applications of the technology, like finding suspects or missing children, can be made safer with laws that create guardrails, like establishing accuracy rates for any software used. The areas that do matter get conflated with broader issues, like whether the police should be surveilling protestors at all, Castro said.</p><p>"We can separate some of these issues without banning the technology," he says.</p><p>CCOPS laws only regulate government use of technology, and don't stop private companies from using it. St. Louis has a handful of businesses, like corner stores, using cameras run by a local company called Blue Line Technology that says it <a href="https://www.kmov.com/news/local-former-cops-create-facial-recognition-technology-to-fight-crime/article_e7694d3a-7689-11e9-ab41-939fb5bbfa86.html">uses face recognition</a>. Another company called Persistent Surveillance Systems (PSS) <a href="https://fox2now.com/2019/07/31/surveillance-company-offers-st-louis-6-6-million-crimefighting-tools-for-free/">offered millions of dollars in crime-fighting technology</a> to St. Louis for free.</p><p>But if the police wanted to plug AI-powered cameras from private property into their Real Time Crime Center or take PSS up on their offer, CCOPS would kick in, forcing the police to bring the project to the Board of Aldermen, points out Marlow. Then the community and its elected officials would decide the future they want to see in their city.</p><p>"CCOPS sets up the conversation; it isn't the conversation itself," he says.</p>
<div class="newsletter"></div>

<p>As law enforcement agencies deploy AI-powered facial recognition systems, some communities are pushing back, insisting on having a say in how they’re used.</p>

A young man operates an Xbox Adaptive Controller, which makes gaming accessible to people with a broad range of disabilities. Photo credit: Microsoft

How People with Disabilities Are Using AI to Improve Their Lives

<p>Kaden Bowen is a smiley 12-year-old who loves music and riding in fast cars. He also can’t walk or talk, is legally blind, and can only use the pinky finger on his left hand.</p><p>Kaden has cerebral palsy, and his father, James Bowen, has been trying out new technology to see if he can create tools that let his son have more independence. One such experiment led James to set up an Amazon Echo Show, a smart assistant device that uses AI to power its voice-control capabilities. Kaden already has a speaking device that has buttons he can press with his working finger to verbalize preselected words or sentences that communicate with the Echo.</p><p>James set it up so his son could ask for things like “Echo, tell me a joke” or “Echo, tell me the news." He also programmed it so Kaden could call his cell phone, Kaden’s grandparents, and a few other people.</p>
<figure class="richtext-image full-width">
<img alt="" class="richtext-image full-width" src="https://www.pbs.org/wgbh/nova/media/images/Kaden_DdPbMRq.width-990_K54qUlj.png"/>
<figcaption><p>Kaden Bowen, 12, who has cerebral palsy and is non-verbal, uses a speaking device to communicate with Amazon Echo. His father, James, is "proud of Kaden’s initiative and grateful for the technology that made it possible." Photo credit: the Bowen family </p>
<div class="credit"><p></p></div>
</figcaption>
</figure>
<p>But Kaden did much more than that. While James and his wife were out one evening, Kaden, who was home with a babysitter, took it upon himself to use his speaking device to activate the Echo, tell it to call his grandfather, and then asked to go out on a car ride—which his grandfather was happy to do. Sharing that story in a <a href="https://www.facebook.com/jsbowen/posts/10214726861239381?__tn__=-R">Facebook post</a>, James said that he was proud of Kaden’s initiative and grateful for the technology that made it possible.</p><p>“It was the first time since he was a toddler playing with a rattler that he was able to interact with something all by himself,” James says. “This Echo device goes way beyond ordering groceries or looking up a recipe for us."</p>
<div class="related"></div>
<h2>Necessity is the Mother of Invention</h2><p>Technology has been opening doors for individuals with disabilities, from motorized scooters to hearing aids, for a long time. And in the coming years, AI will begin to supercharge those efforts with new abilities and expanded access. With one billion-plus people with disabilities around the world, there is plenty of work to be done—and a large market to tap into.</p><p>“They are our customers, our friends, everybody,” said Jenny Lay-Flurrie, the Chief Accessibility Officer at Microsoft, who is herself deaf.</p><p>Disabilities, whether related to vision, hearing, mental health, learning, cognition, or mobility, can be permanent, temporary, or even situational. Designing new products with different levels of abilities in mind—a concept called inclusive design—has gone a long way in ensuring that technology works for everyone.</p><p>Making design widely accessible also ends up being good for those without a disability. One of the first typewriters sprung from a creator’s desire for his blind friend—some say lover—to be able to write more legibly. Alexander Graham Bell’s mother was deaf and <a href="https://www.pbs.org/weta/throughdeafeyes/deaflife/bell_nad.html">his invention of the telephone</a> came out of his work with the deaf community.</p><p>More recently, audiobooks were created as a way for the <a href="https://www.pbs.org/newshour/arts/a-short-history-of-the-audiobook-20-years-after-the-first-portable-digital-audio-device">blind to enjoy literature</a>. Video captioning was invented to make content compliant with the Americans with Disabilities Act (ADA) but now is regularly relied on by anyone watching videos when they can’t listen to the audio.</p><p>Creating new tools could help integrate a segment of our population that has often been left out of routine daily life activities and job opportunities. The unemployment rate is <a href="https://www.bls.gov/news.release/disabl.nr0.htm">twice as high</a> for Americans with disabilities, according to the Bureau of Labor Statistics. Worldwide, <a href="http://www.who.int/en/news-room/fact-sheets/detail/assistive-technology">only one in 10 people</a> who could use assistive products have access to them.</p>
<div class="cove-video">
<iframe allowfullscreen="" class="inline-cove-player" frameborder="0" height="100%" id="partnerPlayer" marginheight="0" marginwidth="0" scrolling="no" src="//player.pbs.org/partnerplayer/INAo3x_BwCaGEBtWspMFxw==/?topbar=false&amp;end=0&amp;endscreen=true&amp;start=0&amp;autoplay=false" width="100%">
</iframe>
</div>
<p>Artificial intelligence hasn’t been around long enough to tackle every opportunity, but Microsoft wants to speed up innovation with a program called “AI for Accessibility” that was announced earlier this year with $25 million in funding. Its goals include developing more AI in the company and giving out grants to others who want to build tools for disabled communities. With AI, Lay-Flurrie says that technology that she thought was five to 10 years out is now more like one to three years away.</p><p>“It’s stupidly exciting for disability, honestly,” she says.</p>
<figure class="richtext-image full-width">
<img alt="" class="richtext-image full-width" src="https://www.pbs.org/wgbh/nova/media/images/Seeing_AI_v2_nIR8n4K.width-990_iQ4Ia1x.jpg"/>
<figcaption><p>Microsoft's Seeing AI app helps the visually impaired by translating text into spoken words. Photo credit: Microsoft<br/></p>
<div class="credit"><p></p></div>
</figcaption>
</figure>
<p>Of course, companies get something out of this, too. Businesses trying to recruit from the limited AI talent pool can use social good projects to convince potential hires that their work will contribute in a meaningful way to the world. Building products that work for previously ignored disabled communities also expands their potential user base. And in a lot of cases, new features can be baked into products already used by non-disabled groups, making them better and more profitable.</p><p>It’s not just big companies innovating in this space. Over 100,000 deaf and hard of hearing individuals have used <a href="https://www.ava.me/">Ava</a>, an app that allows them to take part in group conversations in either English or French (with more limited use for Spanish, Italian, German, and Russian). Everyone engaged in a conversation opens Ava on their phones, then speaks normally as the app listens in. Ava converts spoken words into text in nearly real time, rendering each speaker’s words into a different color for those needing to read along to follow the chat.</p><p><a href="http://www.voiceitt.com/">Voiceitt</a> is an app for people with speech impediments, including both those who need it temporarily after strokes and brain injuries, and those with more long-term conditions like cerebral palsy, Parkinson’s, and Down’s syndrome. Using machine learning, Voiceitt picks up speakers’ unique speech patterns, recognizes any mispronunciations, and normalizes their speech before creating an output of audio or text.</p><p>Products that were designed for a general audience are also being used to increase accessibility. Smart assistants like Amazon’s Alexa and Apple’s Siri have become some of the biggest helpers for blind users, allowing them to get online more easily. Users are discovering new ways to open up voice-based assistants to the deaf, too. One project made an Echo able to <a href="https://medium.com/tensorflow/getting-alexa-to-respond-to-sign-language-using-your-webcam-and-tensorflow-js-735ccc1e6d3f?linkId=55302800">understand and respond in sign language</a> using a webcam.</p><p>“Accessibility is designed into everything today,” Lay-Flurrie says. “We need to make it part of the air that we breathe.”</p>
<div class="ad"></div>
<h2>Driving Innovation</h2><p>Paulo Pinheiro, a 34-year-old robotics PhD from Brazil, was waiting for a flight at the airport when he saw a person his age in a wheelchair who didn’t seem like she could move very well.</p><p>“But she had a great smile,” he says.</p><p>That insight led Pinheiro to launch the startup HOOBOX Robotics. Its first product, called the Wheelie 7, lets wheelchair users initiate moves like going forward, turning, and stopping by making nine distinct facial expressions. While the computer vision behind the technology is cutting edge, the kit itself takes just seven minutes—hence, the name—to add to a wheelchair and is easy enough to install that even non-techie family members can do it. There are currently 55 people in the U.S. using the Wheelie 7—mostly quadriplegics, people with ALS, and senior citizens.</p><p>“Five years ago, this would have been impossible,” Pinheiro says.</p><p>While Pinheiro’s team is still testing the Wheelie 7, which retails for $3,600, he says they are already working on features that would allow users to use facial expression to activate and use smart assistants.</p><p>“Not only can they use it to get around, but they can also use it to connect with the world,” Pinheiro says.</p>
<figure class="richtext-image full-width">
<img alt="" class="richtext-image full-width" src="https://www.pbs.org/wgbh/nova/media/images/34678567211_db56bf9e3d_k_GndroKS.width-990_kIsjapD.jpg"/>
<figcaption><p>The HOOBOX Wheelie uses machine learning to recognize facial expressions to command a user's wheelchair. Photo credit: hoobox2017, Flickr</p>
<div class="credit"><p></p></div>
</figcaption>
</figure>
<h2>An Inclusive Future</h2><p>All types of AI need diverse data sets to prevent algorithms from learning bias or coming up with results that discriminate against certain groups. While this problem is usually invoked in the context of racial and gender discrimination, people with disabilities are also at risk, according to an American Association of People with Disabilities’ (AAPD) technology consultant, Henry Claypool.</p><p>“We need to have a more progressive agenda to make sure people with disabilities are represented in data,” says Claypool.</p><p>Including all types of disabilities will be tricky. Data profiles of people with disabilities are sometimes easy to spot, which makes privacy a concern—especially for conditions that have a high chance for stigmatization, like mental health issues. But gathering the necessary data is doable with the right efforts to anonymize any identifying information. Figuring out how to get this right is essential, Claypool adds: Not being reflected in the data from the start will jeopardize people with disabilities’ access to ubiquitous technologies that are becoming an essential part of the fabric of the modern world.</p><p>“It could have a huge equalizing effect,” Claypool says. But, he adds, “These data innovations won’t benefit everyone if we aren’t careful.”</p><p>James Bowen’s family has felt the effect of the lack of planning for accessibility. An update to the Echo now requires users to confirm that they want to make a call, which makes it harder for Kaden to use. Still, James sees new technology as a way to open up the world for his son and is already researching new tools to try out.</p><p>“It’s the future of what we are looking for and how we are trying to motivate him,” James says.</p>
<div class="newsletter"></div>

<p>Advances in artificial intelligence have spurred the development of smart devices to help people overcome physical and cognitive challenges. And, this may just be the beginning.</p>

Don't expect a robot to treat you at your next doctor's visit. But AI algorithms are playing a bigger role in diagnosis and other aspects of healthcare. Image credit: Getty Images

The Algorithm Will See You Now: How AI is Helping Doctors Diagnose and Treat Patients

<p>In a children’s hospital overlooking the U.S. Capitol, researchers are working to solve a global problem: how to spot genetic disorders in newborns.</p><p>About <a href="https://www.nature.com/scitable/topicpage/birth-defects-causes-and-statistics-863">eight million children</a> are born with a chromosomal abnormality every year, but at least a third aren’t diagnosed until much later. That’s largely because while there are over 6,000 genetic disorders, common newborn DNA tests only look for about 20 of them. Many of those undiagnosed babies experience health complications that end up being responsible for 25 percent of all infant deaths.</p><p>AI could help catch more of these conditions, and faster. The key? About half of all genetic disorders have facial markers—telltale structural features or patterns—which computer vision algorithms can be trained to spot, no matter how subtle.</p>
<div class="related"></div>
<p>At Children's National in Washington D.C., researchers created an application called mGene. The app lets physicians feed a photo of a baby’s face to an algorithm that looks at facial landmarks and takes various measurements like the size of the nose and angle of the eyes, then determines if there is a genetic condition present. So far, their tool can recognize four serious, sometimes life-threatening syndromes—Down, DiGeorge, Williams, and Noonan—with accuracy rates well over 90 percent.</p><p>While mGene is still in its early stages of testing, the researchers behind it are eager to get the tool deployed. Beyond finding cases that clinicians miss, the easy-to-use app can assist in places—all too many of them, unfortunately—that have few or no geneticists. For example, nurses could be trained to use the tool, then direct parents to follow up care as needed, potentially saving lives.</p><p>This research is also expanding what geneticists know about chromosomal abnormalities. mGene is discovering facial landmarks for syndromes previously unknown to doctors, according to Marius George Linguraru, a principal investigator at the Sheikh Zayed Institute for Pediatric Surgical Innovation at Children’s National.</p><p>Importantly, researchers trained mGene on newborns from 20 different countries, making it more likely to spot conditions in a wider range of the world’s population than most practitioners can.</p>
<figure class="richtext-image full-width">
<img alt="" class="richtext-image full-width" src="https://www.pbs.org/wgbh/nova/media/images/mGene_New_Pic_2.width-990_1oMfr1S.png"/>
<figcaption><p>mGene, an app created by researchers at Washington D.C.'s Children's National, analyzes infants' facial landmarks, allowing physicians to identify and diagnose genetic conditions.</p>
<div class="credit"><p></p></div>
</figcaption>
</figure>
<p>“Geneticists are mainly trained on textbooks with examples of northern European ancestry,” said Linguraru. But that doesn’t always work, he adds, since humans look so different from one another, both within and among ethnic groups. For example, certain eye shapes can be a sign of Down syndrome in white babies, but are not a meaningful signal in Asian populations.</p><p>The team is also working on gathering more data to train mGene on other genetic disorders. Recognizing thousands of conditions—many of which are so rare that there are only a handful of cases in the world—is impossible. AI could fill in the gaps.</p><p>“There is only so much that a clinician can look at and understand,” said Antonio R. Porras, a researcher working on mGene.</p>
<div class="ad"></div>
<h4>From Data to Diagnosis</h4><p>More than most other fields, AI has been put to use in healthcare to see if it can find new solutions to old problems in the data. So far, it has shown promise in improving medical procedures, expanding services, and finding novel ways of offering care.</p><p>As much as AI is bound to change the industry, however, doctors and nurses will remain a crucial part of the healthcare experience. Daniel K. Sodickson, a radiology professor at New York University working with Facebook on <a href="https://medium.com/syncedreview/fair-nyu-school-of-medicine-share-fastmri-tools-release-largest-ever-mri-dataset-b240247031bf">a project to speed up imaging from MRIs</a>, says the most prevalent misperception that he encounters is that a doctor’s job is to look through the data and find the problem.</p><p>“It’s a very small part of what physicians do,” Sodickson says.</p><p>Analyzing MRI images has proven to be a particularly productive area of AI research. Again and again, technologists have shown that algorithms are able to spot abnormalities in X-rays as well as, if not better than, radiologists. In a now infamous <a href="https://www.newyorker.com/magazine/2017/04/03/ai-versus-md"><i>New Yorker</i></a> story, computer scientist Geoffrey Hinton, one of the people most responsible for the current AI boom, was quoted saying that “we should stop training radiologists right now.”</p><p>But the narrative is evolving. The actual AI being applied to real-world problems is acting more like a talented assistant than it is auditioning for the role of a new, improved robotic doctor.</p><p>“Replacing the human doctor is not our business,” says Jeff Nadler, the CIO for Teladoc Health, a virtual health provider using AI to advance their systems.</p><p>With its telehealth services, Teladoc saves money for patients and gets them in front of a doctor faster, says Nadler. Machine learning reduces wait times and decides how many doctors need to be available at any given time. As a result, telehealth can create more ways for clinicians to see patients and shorter wait times, removing common barriers to getting healthcare.</p><p>“We're creating a more long-term relationship between the clinician and patient,” Nadler said.</p><p>The power of combining human and artificial intelligence is becoming evident in other research as well. In 2016, a team of pathologists at Harvard created an <a href="https://hms.harvard.edu/news/better-together">AI-powered program</a> to identify breast cancer cells. Pathologists accurately spotted breast cancer 96 percent of the time, beating the algorithm’s rate of 92 percent. But the combination of the two found 99.5 percent of cancerous cells. With nearly 1.7 million new cases of breast cancer diagnosed every year, as many as 130,000 more patients could receive accurate diagnoses through such collaborative efforts between humans and computers.</p>
<figure class="richtext-image full-width">
<img alt="" class="richtext-image full-width" src="https://www.pbs.org/wgbh/nova/media/images/1280px-Breast_cancer_cells.width-990_paMv9Ay.jpg"/>
<figcaption><p>In a recent study, AI algorithms helped pathologists identify breast cancer cells (like those pictured here) in lymph node tissue samples with 99.5 percent accuracy, up from 96 percent. Image credit: National Cancer Institute, Wikimedia Commons</p>
<div class="credit"><p></p></div>
</figcaption>
</figure>
<h4>Wrestling with the Ethics of AI in Medicine</h4><p>The field of healthcare seems uniquely poised to deal with problems that have plagued other industries—like privacy and ethics—that have seen an AI revolution. HIPAA, or the Health Insurance Portability and Accountability Act, is American legislation that outlines data privacy for medical information that has been in place and guiding healthcare since 1996. The <a href="https://www.health.harvard.edu/blog/first-do-no-harm-201510138421">Hippocratic Oath</a> that all physicians take, stating they will “do no harm,” also adds a layer of consideration for doctors looking to use AI. Because of the industry’s stringent rules and how much research is being done in healthcare, providers might end up establishing best practices for uses of AI in <i>all</i> fields, not just medicine.</p><p>“We’re in this wild west phase with AI right now, everyone is trying everything,“ Sodickson says. “We need a sheriff’s office.”</p><p>With so many applications and more in the pipeline, the integration of AI in healthcare will not be easy. There are practical considerations, including the technical act of rolling AI out. It won’t be just one or two algorithms in play, but hundreds or thousands that could possibly disagree, according to Parsa Mirhaji, the Director for the Center for Health Data Innovations, and CTO of NYC Clinical Data Research Network.</p><p>“It becomes a headache. Which one do you believe and why?” he says.</p><p>Mirhaji and his team are working to mentor and train healthcare professionals on how to use and think about AI. But it's going to be a sea change that will take some time to sort out, he stresses. For all the help that AI can bring, integrating it will bring growing pains that humans will have to figure out on their own.</p><p>Mirhaji concludes: “I am not aware of an AI algorithm I can build that will fix that.”</p>
<div class="newsletter"></div>

<p>Artificial intelligence researchers are building tools to quickly and accurately turn data into diagnoses. But practical limitations and ethical concerns mean humans should remain in charge.</p>

These dancers from a Syracuse-based dance studio helped with the design and branding of the AI-powered chatbot Layla. Image credit: Tiffany Lloyd, Layla

With ‘Problems Out There Waiting for Us,’ Public Health Practitioners Turn to AI for Help

<p>When Yanira was a teenager, she didn’t feel like she could ask her parents about sex. So like most kids her age, she would try looking up answers to her questions online—only to get stuck looking up the medical terms some sites used, or trying to sort out whether the source was legitimate or not.</p><p>Now Yanira is 20 years old, and her 15-year-old sister is turning to <i>her</i> with questions. But Yanira doesn’t have to Google anything anymore: She sends her sister to a chatbot named <a href="http://laylasgotyou.com/">Layla</a>, an AI-powered tool built by a public health nonprofit for the girls of Syracuse, New York.</p><p>“It was a lot better than I could have thought of off the top of my head," she says of the information provided by Layla.</p><p>Yanira knew about the site because she had been part of a group of young women to give feedback on the chatbot. She liked the way it responded to questions, and thought it made answers clear. She also liked that it provided links to where to get real-world services in her area.</p>
<div class="related"></div>
<p>“When I was 15, I didn’t know much about birth control, much less where to get it," Yanira says.</p><p>Teen pregnancy is a real problem in Syracuse. For girls between ages 15 and 17, the rate is 21.8 per 1,000, compared to 18.8 in the U.S. as a whole. And for black and Hispanic teenagers, the rates are much higher: 65.6 and 53.3 per 1,000, respectively. A study found that 80 percent of those pregnancies were unplanned.</p><p>That same research, conducted by Public Good Projects (PGP), the public health nonprofit behind the chatbot, found that a major cause behind the high pregnancy rates was these young women’s limited understanding of sex.</p><p>“What does a condom look like? They couldn’t tell you," says Joe Smyser, the CEO of PGP.</p><p>Finding useful, accurate information can be hard for teens. Random internet searches run the risk of finding incorrect information, while reputable sources can be written in dry language that doesn’t reflect the lived experience of teenage girls. Going to parents isn’t always an option, either.</p><p>But Layla is available 24/7, and no question is off-limits. As the opening prompt says, Layla wants to be considered “your new best friend with all the deets about those uncomfortable topics like birth control, sex, and STDs.”</p><p>PGP enlisted a group of young women from Syracuse to help them design the chatbot, which uses IBM’s Watson technology—specifically its natural language processing abilities—to make the chatbot nimbler at answering questions phrased in a variety of ways. Although cutting edge AI powers the site, everything from the resources it recommends, to the language it uses, to its aesthetic look and feel were designed by local teenagers who need it most.</p><p>“We didn’t want this to come off like a healthcare department,” says Tiffany Lloyd, Layla’s community manager. “It was created by brown girls, for brown girls.”</p>
<figure class="richtext-image full-width">
<img alt="" class="richtext-image full-width" src="https://www.pbs.org/wgbh/nova/media/images/Unknown_MbIH13c.width-990_D6C2Bbr.jpg"/>
<figcaption><p>Jason Wagner, 16, and Deziah Thomas, 15, are two youth influencers that helped to promote Layla, a chatbot designed to help young people of color get reliable information about sexual health, including where to find additional resources and services. Image credit: Tiffany Lloyd, Layla</p>
<div class="credit"><p></p></div>
</figcaption>
</figure>
<h4>AI for the Globe</h4><p>The field of public health has the ambitious goal of preventing the spread of diseases and extending life across human populations that span everything from small communities to entire continents. It’s a daunting challenge with new problems constantly cropping up, often with insufficient resources to tackle them. Those challenges have led public health workers to look to AI to find novel ways to solve global health issues like tuberculosis and teenage pregnancy.</p><p>One of the earliest and best-known attempts to use big data for public health—done before AI research really took off—was <a href="https://en.wikipedia.org/wiki/Google_Flu_Trends">predicting the spread of flu from Google searches</a>. While Google has ended its internally run research, the company has added new AI capabilities and still passes data along so that places like <a href="https://www.sciencedaily.com/releases/2019/01/190111143744.htm">Boston Children's Hospital can continue the work</a>. By going through electronic health records, flu-related Google searches, and historical flu activity in a given location, the AI-powered tool has demonstrated the ability to spot flu outbreaks two weeks before the Centers for Disease Control (CDC) can.</p><p>This approach is not entirely new. Disease surveillance systems now routinely track people researching symptoms in Google searches, as well as more traditional, expensive-to-gather clinical data. But more sophisticated data crunching algorithms exist for marketers: Layla developer PGP, for example, used a tool originally designed to monitor “brand health,” a mix of various metrics companies look at to see how their company is perceived by the public, and repurposed it for tracking opioid use and mental health trends.</p><p>“No one in academia understands Twitter the way marketers do,” Smyser says.</p><p>With data from a wide range of traditional and social media sources, including closed captioning from local television, newspapers, Facebook, Twitter, Instagram, and YouTube, PGP is generating new insights into opioid use. But it’s only possible with natural language processing, a subset of AI that parses written or spoken word—like the Layla chatbot does with IBM’s Watson. In this instance, the AI learns to find stories and tweets about opioids, even if slang is used, without having to be taught every different way the drug use might be referenced.</p><p>Right now, the system picks up about 55,000 references to opioids a day. These might include everything from a tweet from <a href="https://twitter.com/realdonaldtrump/status/1091127643720859649?lang=en">President Trump boasting of a fentanyl bust</a> to Facebook posts of dealers hawking a product. With such a large stream of data, experts are able to conduct a lot of previously out-of-reach analysis. Smyser is sharing what PGP is learning with presentations to agents from the Drug Enforcement Agency (DEA) and community groups, but there are limits to how much small, community-based nonprofits like his can help.</p><p>“There is no red phone for me to pick up and tell CDC they need to be aware of this,” Smyser says.</p>
<div class="ad"></div>
<h4>Old Methods, New AI Insights</h4><p>Other public health AI projects don’t rely on internet data. There are nearly 2 million homeless young people between 13 and 24 in the United States, so researchers out of the University of Southern California’s Center for AI in Society (CAIS) developed software called <a href="https://www.cais.usc.edu/projects/predictive-models-of-vulnerability-and-housing-prioritization/">TAY Triage Tool</a> to combat the problem. It uses information from a short questionnaire to predict which young people are at risk of long-term homelessness, allowing decision-makers to divert precious resources to those most in need.</p><p>A similar tool called <a href="http://teamcore.usc.edu/people/amulya/healer/index.html">HEALER</a> figures out who in a group of homeless youth is the best person to train about HIV prevention. The traditional approach is to bring in and train the most popular person. But HEALER actually discovers different members of the group with more widespread connections than the most popular individuals, who typically hang out with smaller groups.</p><p>“We are trying to train gatekeepers to watch over others in their social networks,” says Milind Tambe, founding co-director of CAIS and an engineering professor at USC, who worked on both projects.</p><p>Sometimes, AI algorithms just confirm best practices and the cost-effectiveness of solutions, always a priority for cash-strapped public health programs. There are over two million individuals with tuberculosis in India, and treating it presents a major challenge: Multi-drug resistant (MDR) tuberculosis is widespread and hard to diagnose at scale.</p><p>“Finding these strains of tuberculosis is really important,” says Sze-chuan Suen, a professor of engineering at USC.</p><p><a href="http://bit.ly/2OmYA1b">GeneXpert</a> is a test that can detect drug-resistant bacteria—but it does at a high cost. Suen did a study on this expensive screening and concluded, despite GeneXpert being costlier in the short term than other methods, it should be used at initial diagnosis for all patients. The benefits of identifying patients with drug-resistant strains of TB should save money—and lives—in the long run.</p><h4>Part of a Balanced Global Diet</h4><p>Compared to industries like healthcare or education, public health has not had a massive influx of AI projects. In part, that’s because much of the progress in AI has been around image recognition, which can be most readily applied to individual medical diagnoses. For example, Stanford researchers reported in 2017 that an artificial neural network they built could diagnose dangerous melanomas from photographs as accurately as experienced dermatologists, which could expand access to care in areas with fewer specialists.</p>
<div class="cove-video">
<iframe allowfullscreen="" class="inline-cove-player" frameborder="0" height="100%" id="partnerPlayer" marginheight="0" marginwidth="0" scrolling="no" src="//player.pbs.org/partnerplayer/mBsCawbZnOV78ixQxlxyIg==/?topbar=false&amp;end=0&amp;endscreen=true&amp;start=0&amp;autoplay=false" width="100%">
</iframe>
</div>
<p></p><p>Another factor limiting AI applications is the small budgets and profits to be made in public health, which lowers the likely return on investment in new technologies. And many public health projects go through ethics boards that determine the rules around use of public data and research protocols. If anything, these processes “can be cautious to a fault” and delay approval, according to Henry Kautz, a computer science professor at the University of Rochester.</p><p>That doesn’t mean AI won’t be utilized in public health in the near future in a significant way. The problems are simply too big, and AI’s potential too enticing, not to try.</p><p>“There are problems out there waiting for us,” Tambe says.</p><p>Another challenge in applying AI to public health is the field’s focus on whole populations, not individuals. Unlike other industries that create tools that individuals can put into use—such as <a href="https://www.pbs.org/wgbh/nova/article/does-ai-hold-key-to-new-green-revolution-in-agriculture/">apps for recognizing farm pests</a> or <a href="https://pbs.org/wgbh/nova/article/ai-technology-is-disrupting-the-traditional-classroom/">personalized learning tools</a>—public health demands solutions for broad populations with an inherently wider set of needs. This often means that data and algorithms can only take us so far before humans must get involved.</p><p>Take foodborne illnesses: AI algorithms can sift quickly through large datasets to spot problems, but that’s only half the battle—government inspectors must then ground truth the findings with in-person inspections.</p><p>For the chatbot Layla, that means providing phone numbers to real people for further help for search queries that require follow up, like where to get STD testing or services for sexual assault victims. Lloyd, Layla’s community manager, says that more outreach work needs to be done to understand and address issues that are too nuanced for current AI.</p><p>“Make no mistake, the chatbot is not the end-all,” Lloyd says.</p><p>Lloyd continues to connect with the young women Layla is meant to serve in focus groups, and makes sure the word gets out with social media campaigns and posters at places like hair salons.</p><p>“They need an ongoing conversation,” Lloyd says.</p><p>Still, for underserved populations like teenage girls in Syracuse, AI like Layla has its place and has room to grow. Yanira had suggested other capabilities for the site, like a period tracker. In the meantime, Yanira says she would recommend it to friends and even her two other little sisters.</p><p>“It is kind of like another big sister,” she says.</p>
<div class="newsletter"></div>

<p>Scarce funding, ethical concerns, and other problems have limited the use of artificial intelligence to tackle public health issues. That’s starting to change.</p>

Over 2 billion people are living in areas of water scarcity, according to the United Nations, and that number could grow to 5 billion by 2050. Implementing AI could help. Image credit: iStock, brazzo

Thirsty for solutions, water managers are putting AI-powered tools to work

<p>When Meena Sankaran was a child growing up in Mumbai, India, her family didn’t have consistent access to water in their home, and what they did have wasn’t always safe to drink. She was often sick—with typhoid, malaria, jaundice, mumps, and pneumonia, all before she was 17. If these illnesses weren’t directly water-related, the lack of clean water didn’t help her recovery.</p><p>“My immune system just got used to it,” Sankaran says.</p><p>Sankaran is one of the lucky ones. The <a href="http://documents.worldbank.org/curated/en/586371495104964514/pdf/115133-WP-P152203-PUBLIC-17-5-2017-12-28-1-WaterlifeCaseApril.pdf">World Bank estimates</a> that almost a quarter of all communicable diseases in India are due to unsafe water and the lack of safe hygiene practices, with more than 500 children under the age of five dying each day from diarrhea.</p><p>Sankaran spent 20 years getting an engineering degree in the U.S. and working in the tech industry before using her personal experiences to launch a startup in 2016 called KETOS, a water management company that uses AI to monitor and test quality.</p>
<div class="related"></div>
<p>The company has two products: KETOS Wave, a device that sits in a water pipe and can detect more than 20 parameters like flow, pressure, usage, and volume that feed into a machine learning algorithm that can identify possible leaks and track use. So if water volume goes down in a way that the algorithm parsing the data doesn’t expect, Wave can notify the user of the likely leak with a mobile app, which also has the ability to remotely shut the water off.</p><p>The company’s other product is KETOS Shield, which can measure and detect 20 different dangerous contaminants like lead, copper, nitrates, arsenic, and chlorine. Measuring these toxins doesn’t utilize AI, but Shield combines routine chemical testing with factors that can contribute to fluctuations—such as the time of year, or the temperature—in a machine learning algorithm that predicts chemical spikes before they happen.</p><p>While the 140 deployments to date are mainly with municipal and agricultural clients in the U.S. and Mexico, Sankaran also wants to do something for her homeland. That led KETOS to offer water solutions across India as part of the government’s <a href="https://www.smartvillagesinitiative.org/">Smart Village Initiative</a>. In 2017, the company deployed its technology in 25 villages. None of the villages had a way to monitor quality, and several could go days without getting water at all. Sankaran said they were able to improve that supply by up to two hours a day in most cases.</p>
<figure class="richtext-image full-width">
<img alt="" class="richtext-image full-width" src="https://www.pbs.org/wgbh/nova/media/images/IMG_7090.width-990_fHrHYXR.jpg"/>
<figcaption><p>Meena Sankaran (second from left), is the founder of the tech startup KETOS, a water management company that uses AI. In 2017, KETOS deployed its technology to 25 villages in Sankaran's homeland, India. Image credit: KETOS</p>
<div class="credit"><p></p></div>
</figcaption>
</figure>
<p>Not that there haven’t been hiccups. Saravana Sarru is a resident of Kuppam, the first village in the State of Andhra Pradesh that KETOS worked in, and ended up working as a liaison on the ground for many of the other villages. In some of them, the rainy season interfered with the devices sending information back to those monitoring the water, a problem that was solved by relying on a mesh network, a type of network that made internet connectivity more reliable. Even with the initial issues, Sarru said that all the communities have been excited by the potential to have more control over their water source.</p><p>“It’s very new to India,” Sarru says.</p><p>KETOS is planning a project in Majuli, the world’s largest river island, which has major problems with arsenic contamination. They are actively partnering with an Indian NGO to recruit local women as field support personnel, giving them jobs that are also empowering villagers to take care of their own water. Thousands of other villages have expressed interest, but the recent elections have halted the project.</p><p>“We are slowly but surely pressing forward,” Sankaran says. “India takes time.”</p><p>But the country has no time to spare in finding new ways to better monitor and improve water management. More <a href="https://washmatters.wateraid.org/publications/the-water-gap-state-of-the-worlds-water">than 163 million Indians</a> don’t have access to clean water, the highest number of any country in the world, and more than 600 million face acute water shortages. Nearly two dozen cities could run out of groundwater by next year, including megacities like Bangalore and New Delhi.</p>
<div class="ad"></div>
<h2><b>A Global Problem</b></h2><p>Across the rest of the world, over 2 billion people are living in areas of water scarcity, <a href="http://www.unwater.org/water-facts/scarcity/">according to the United Nations</a>, and that number could grow to <a href="https://www.theguardian.com/environment/2018/mar/19/water-shortages-could-affect-5bn-people-by-2050-un-report-warns">5 billion by 2050</a>. Nearly a million people die each year from water, sanitation, and hygiene-related diseases that would be reduced with access to clean water.</p><p>The U.S. has one of the safest overall systems but still has serious issues to contend with in the coming decades. America’s drinking water is delivered through more than a million miles of pipes, most of which were laid in the early to the mid-20th century with a lifespan of 75 to 100 years. This past-retirement age system contributes to an estimated 240,000 water main breaks each year—and some two trillion gallons of water lost annually. The government will need to spend a trillion dollars maintaining and expanding services for a growing population over the next 25 years, according to the American Water Works Association.</p><p>These old pipes also create water quality problems. While the bigger public drinking water systems have the ability to test water frequently, smaller municipal systems that still serve millions of Americans in more rural areas don’t necessarily have the budget or expertise to do the same amount of testing. AI-powered tools like KETOS could help bridge the gap, and identify problems that otherwise might remain invisible for years.</p><p>“We are desperately in need of better monitoring,” says Peter Gleick, a climate and water scientist who co-founded the Pacific Institute. “It’s a problem, even here in the U.S.”</p><p>By far the most public water crisis has been unfolding in Flint, Michigan, where over 100,000 citizens have been exposed to dangerous levels of lead leaching from their pipes due to insufficient treatment. The state has struggled for five years to make it right. AI might be part of the solution to finally helping Flint: Volunteer researchers <a href="https://www.theatlantic.com/technology/archive/2019/01/how-machine-learning-found-flints-lead-pipes/578692/">developed a machine learning algorithm</a> to predict which homes are most likely to have lead pipes that need to be changed out.</p><p>Implementing this inspection system has proven difficult, however. Some residents whose pipes were not prioritized by the algorithm have complained, feeding into the general atmosphere of <a href="http://bit.ly/2IPETgm">distrust in politicians and policymakers</a> that still simmers in Flint five years after the crisis started. Roughly one-quarter of the water pipes in the city have yet to be inspected, and even when that’s done and any remaining lead pipes have been replaced, an unknown number of damaged pipes <i>within</i> homes will remain, ensuring that Flint’s struggle for safe drinking water will drag on for years to come.</p>
<figure class="richtext-image full-width">
<img alt="" class="richtext-image full-width" src="https://www.pbs.org/wgbh/nova/media/images/Sequence_01.00_38_26_19.Still001.width-990_IJXhwSo.jpg"/>
<figcaption><p>Activists and residents gather at a water treatment plant in Flint, Mich. to commemorate the 5-year anniversary of the Flint water crisis. Image credit: Jason Vlahos, WKAR</p>
<div class="credit"><p></p></div>
</figcaption>
</figure>
<p>The problem isn’t unique to Flint, however. There are an estimated <a href="https://www.cbsnews.com/news/lead-in-americas-water-systems-is-a-national-problem/">6.1 million lead pipes across the country</a>. Flint is the only city that has so far been able to prove a causal link between its water and elevated lead levels, but a <a href="https://www.reuters.com/investigates/special-report/usa-lead-testing/">Reuters’ investigation</a> found 3,000 areas in the U.S. with lead poisoning rates that were at least double those that were seen in Flint during the height of that crisis. Most spots have what the report calls a “lead legacy” made up of old pipes, abandoned industrial waste, and peeling paint that could all be contributing to elevated lead levels.</p><p>“We’ve known for a long time that there are certain water problems that require a lot of data and observations to get a hand on,” Gleick says. “If AI can help sort through that data, it would be enormous step forward in better management of our limited water resources."</p><p>Sankaran’s work in India got KETOS more business but she said that’s not the point. She wants to see a focus on using new tools, like the ones she’s developed, to make the world a better place.</p><p>“You have to question yourself about what’s the benefit of all these advancements in technology if you can’t improve the quality life at a basic level,” Sankaran says.</p>
<div class="newsletter"></div>

<p>Around the world, aging and inadequate water systems are a huge public health problem. Now, researchers are using artificial intelligence to help conserve and monitor the quality of drinking water.</p>

Jackie Snow

Articles