Emails take too long. Well, at least according to Google, which is why on Tuesday at Google I/O, the company announced a new Gmail feature called Smart Compose. The feature allows for artificial intelligence to auto-fill information in the emails you compose in an effort to cut down on the time spent typing up mindless messages. If typing emails is truly one of the bane’s of your day-to-day existence, this feature is here to alleviate that stress.
Smart Compose builds on the Smart Reply feature that Google added to Gmail’s mobile app last year, which would offer three context-appropriate responses to emails. Smart Compose takes that to another level by potentially allowing one to write full emails and thoughts right alongside AI-assisted phrasing instead of just deferring to single choices being offered up.
To use Smart Compose, you’ll simply hit the tab button to add the auto-fill text to your message, which appears automatically as you type.
The feature is scheduled to roll out in the next few weeks, but at least some users can try it now. Just click “Try the new Gmail” under settings, if you haven’t already. Then click Settings, and under General, select “experimental access,” which will enable Smart Compose. (This feature wasn’t currently available in our Gmail, so we haven’t had the chance to test it out yet.)
Personally much of the hassle I experience with emails doesn’t start with composition but rather the internal debate over whether to respond to certain emails or even whether to open them at all. Does Smart Compose help with the existential dread of answering an email at 8:53pm on a Tuesday night or wonder how fast is too fast to reply to an email over the weekend? Perhaps by next Google I/O, its AI will be able to address that problem.
The LAPD Uses Palantir Tech to Predict and Surveil ‘Probable Offenders’
https://ift.tt/2jKcOee
Analysts with the Los Angeles Police Department are reportedly using Palantir software to direct officers to surveil “probable offenders” throughout the city, many of whom are not criminal suspects but have been spotlighted by the company’s predictive technology, according to LAPD documents.
In Justice Today reviewed internal LAPD documents from October 2017 that point to a persistent surveillance campaign compelling analysts to maintain rotating lists of targets selected by agency data-mining techniques and predictive policing tech pioneered by Palantir. The documents were obtained through a public records request filed by the Stop LAPD Spying Coalition, which provides them to In Justice Today.
The In Justice Today report points to what the LAPD calls Chronic Offender Bulletins. Essentially, these bulletins are profiles, partially generated by Palantir software, for individuals who have had some contact with the LAPD. Violent crimes, gun crimes, suspected gang affiliation, and other designations increase a person’s Chronic Offender Score. A high score alone is not enough to justify detainment, but officers are given one-page summaries of a person’s arrest history, notable physical features (referred to as “physical oddities”), cars they own, and a list of where they’ve been stopped by police.
The LAPD used Chronic Offender Bulletins before Palantir’s involvement, but the report notes that the process is both increasingly automated by its software and per-infraction penalties are more severe. A spokesperson for Palantir disagreed with this characterization, telling In Justice Today that the CBO creation is “a human-driven process,” although Palantir software is used in the creation of CBOs, the spokesperson said.
As the report notes, a feedback loop emerges: the LAPD targets those with high scores for increased surveillance, but each stop by police further increases their score. Troublingly, analysts are directed to create a minimum of 12 Chronic Offender Bulletins, with five to 10 “back ups” to be switched in as people are arrested. To be removed from the list, an individual has to go two years without contact—a near impossibility if officers are being compelled to make constant contact with them. The LAPD tracks the number of high scoring “offenders” arrested, and officers are expected to report on COB arrests at weekly meetings, In Justice Today found.
The LAPD did not immediately respond to our request for comment.
AI can help a billion people, but Microsoft can’t do it alone
https://ift.tt/2FZ7fkY
Getty Images/iStockphoto
“It cracks me up when I meet someone who says, ‘Hey, I don’t think I have people with disabilities in my company.’ And that’s when I know they’ve got people there that are not speaking up.”
I’m sitting with Jenny Lay-Flurrie, Microsoft’s chief accessibility officer. We’ve bogarted a few chairs in a hallway at the Washington State Conference Center in Seattle, where the company is holding its annual Build developer conference. Lay-Flurrie, who is deaf, speaks with a crisp English accent. Though she never signs when addressing me, she has brought a sign language interpreter to clarify the questions I’m asking, in case lip reading ever falls short. Together, we are among the few women I’ve seen at this tech convention.
None of this would normally be relevant — not our genders nor Lay-Flurrie’s deafness — except in this case it is. For Microsoft, a giant in an industry not known for its inclusiveness, people with disabilities represent the next chapter in a quest to achieve equality.
Less than three hours earlier, Satya Nadella closed Build’s day-one keynote not with coding demos but some remarks about humanity. The company plans to use artificial intelligence to help people with disabilities, he said, and will be investing $25 million in grants over five years. Throughout the keynote, closed captioning ran across the big screens in the auditorium.
This wasn’t just a gesture of goodwill — it’s also good business. More than 1 billion people, or roughly 15 percent of the world’s population, have a disability, according to the World Bank, encompassing what has to be one of the world’s largest underserved markets. Disability is an umbrella term that includes lifelong impairments and temporary illnesses, including both physical ailments and mental health disorders. Some people are born with disabilities; many more acquire them later, often in old age.
“Disability is thus not just a health problem,” the World Health Organization concludes in its definition of the term. “It is a complex phenomenon, reflecting the interaction between features of a person’s body and features of the society in which he or she lives.”
The stakes are high indeed. According to a report from the US Department of Labor last summer, the unemployment rate for people with disabilities was 10.5 percent in 2016, compared with 4.6 percent for people without disabilities. Only one in 10 people with disabilities have access to assistive products, according to WHO.
Microsoft’s Seeing AI app for iOS uses computer vision to do things like see currency, read handwriting and speak text.
Microsoft, too, has been busy. Last summer the company introduced the Seeing AI app for iOS, which uses computer vision to read handwriting, identify currency, speak text and even suss out the emotion in people’s facial cues. Soon the OS will get a better-organized Ease of Access settings, with the ability to navigate menus using Narrator. Separately, Microsoft is testing eye control for users with limited mobility.
The benefits of AI go beyond accuracy and speed too: It’s also cheap. Both Seeing AI and Microsoft’s Translator app are free to download. And while it might be a stretch to envision a world where AI replaces Seeing Eye dogs, these tools might flip that 1-in-10 ratio of those who are without access to assistive tools.
Microsoft Accessibility team member Anne Taylor
All told, the company has several ways of sourcing feedback from the community it’s trying to reach. There are Microsoft employees with disabilities, user forums, focus groups and the fantastically acronymed Disability Answer Desk (DAD!), which doubles as a dedicated technical-support line and a way for Microsoft to collect common complaints. And yet Lay-Flurrie insists that “no one would ever say we have enough feedback from people with disabilities.” In part, that’s because the ease with which someone can rate or review a product is itself a measure of how elegantly designed it is. “We desperately want more. We’re trying to make it easy and streamlined for people to give us feedback.”
But according to Lay-Flurrie, who has been with Microsoft in various roles since 2005, the company first had to rethink its own office culture. “Not just me, but a lot of people have helped to drive it over the time,” she said. “We’ve been doing this stuff internally for quite some years. Getting our house in order and really working on that. And we’ve got a lot more to do.”
If it seems that change has been coming faster lately, though, that’s because it has. Lay-Flurrie says that when she entered the workforce some 20 years ago, closed captioning was in use, but had yet to fulfill its promise. “I had somebody tell me that closed captioning was going to be a reality,” she recalls. “Twenty years later, there’s really not been much tangible progress to consumer-based products. AI has completely jumped that in the space of the last 18 months. Twenty-two years, I’ve been dogfooding this product to the point where it’s accurate enough I can use it.”
For Lay-Flurrie, that means few, if any, inaccurate captions. “For someone who relies on captions, someone who’s deaf, that inaccuracy means it’s not usable.”
Even by the metric of accuracy, there’s still work to do. “We are so not done,” Lay-Flurrie added. “Not every video has captions. And not every website is successful, not every product — you can’t walk into a supermarket right now and have a fully inclusive experience.” Indeed, even lower-tech solutions can be hard to find in some parts of the world. Just last month, The New York Timesreported that more than a billion people who need eyeglasses don’t have them. Why should we expect AI or even closed captioning to become ubiquitous anytime soon?
Then there are other problems that remain entirely unsolved. For Lay-Flurrie, the holy grail is converting sign language to text. She also sees an opportunity for AI to assist people struggling with mental health problems like anxiety and post-traumatic stress disorder. Perhaps AI could track when you need drugs, she says, or recognize when you’re heart rate is up or when you’re flushed. “It knows the right things and the steps to take when for whatever reason you’re not able to.”
It’s a compelling idea — now someone needs to build it. “We haven’t done a ton yet. I’ve got nothing to crow about,” Lay-Flurrie said. “It’s more a massive opportunity for AI, because there’s so much knowledge.” But it likely won’t be Microsoft executing on the PTSD idea. “We know we’re at the tip of the iceberg,” she added. “It really is democratizing, seeing where else we can drive impact more broadly. And the best and easiest way to do that is to empower others. Empower the developer crowd.” Specifically, the AI for Accessibility program calls for a mix of grants, seed investments and expert advice, when it makes sense.
Microsoft has also been working with some of the same titans it normally calls its competitors, including Apple, Google and Engadget’s parent company, Oath, among others. Accessibility advocates from these firms share best practices and frequently end up sitting next to each other on industry panels. “What grounds us is things like the unemployment rate,” Lay-Flurrie said. “You can’t have an unemployment rate that’s double and then think this is a really big compete space. There’s so much to do. Too much to do for one company alone.”
Surely, Microsoft will continue its efforts after this five-year program ends. $25 million isn’t even a lot of money by the company’s standards. Other than a commitment to accessibility, though, does Lay-Flurrie know what’s next? “Oh no,” she said. “The world was moving this way five years ago, but I don’t think any of us anticipated quite the acceleration that we’d be in at this point.”
At this point, Lay-Flurrie is reminded of something Nadella said in his keynote: the world is becoming a computer. Technology will just be that ubiquitous. “The one vision that keeps me grounded, that I think is going to be incredible, is if you imagine a table full of kids with multiple different disabilities,” she said, “and instead of having to leave the classroom or use very big or bulky or specialized equipment, they’re using the same equipment as anyone else, but in their own way. It’s just mapping the technology to the individual as opposed to the other way around.”
Google Lens will be available in stock camera apps
https://ift.tt/2KIskDE
AOL
Google has been busy updating Lens, its AI-powered image recognition tool, over the past year. It can recognize dog and cat breeds, is available on iOS, and on non-Pixel Android phones. Now, at I/O, Google is rolling out the latest Lens update: It will now be integrated directly into the stock camera app. It’ll start with the Pixel, but will also roll out to other Android phones like the recently announced LG G7.
The company is also rolling out several new features to Lens: smart text selection, style match, and real-time results. With smart text selection, you can do things like copy-and-paste text from the real world directly into your phone, and even quickly finding the meaning of words from a document. In a demo on stage, for example, Google showed how you could quickly tap around words on a menu, and Lens will tell you what exactly that food is, complete with visual guide and an ingredient list. Which is especially useful if the menu is in a foreign language you don’t quite understand.
Developing…
Click here to catch up on the latest news from Google I/O 2018!
Google Maps unveils its first-ever augmented reality interface
https://ift.tt/2I1dkD0
Instead of relying on an unclear blue dot while walking around a cluttered city, Google suggests this camera-fueled augmented reality interface may one day save the day.
Google
Brief tease of business information attached to anything in your direct view while using this new Google Maps mode.
There’s no release date at all whatsoever, and yet there’s already an animal assistant. Hi, fox.
Google used its Tuesday I/O keynote to unveil a pretty killer new feature that may one day come to Google Maps: camera-assisted walking navigation.
“Here’s how it could—will look like,” Google VP Aparna Chennapragada told the I/O crowd when unveiling a sample interface that combines Google Maps’ 2D interface with the view from your smartphone’s camera lens. A small semisphere of map data appeared at the bottom of the interface, while the camera perspective included bold images of where to turn and go—and floating panels that show information about businesses in your direct view.
Chennapragada offered nothing in the way of a release date.
in the U.S. are climbing at an alarming rate, jumping 46 percent since reaching a low point in 2009, according to federal data. Now, a new study from the
‘ findings: Pedestrian crashes have become not only more frequent, but deadlier, with deaths per 100 crashes with pedestrians rising 29 percent from 2010, when they reached their lowest point, to 2015. Unsurprisingly, the increase is happening mostly in urban and suburban areas, in the dark, and the fatalities are generally happening away from intersections, on busy main roads or arterial roads.
It’s the last finding that might be especially stinging for the American buying public: Fatal pedestrian crashes are increasingly likely to involve
or high-powered vehicles, as measured by the ratio of horsepower to weight. IIHS says while pedestrian crashes still most frequently involve cars, fatal single-vehicle crashes involving SUVs increased 81 percent, more than any other type of vehicle.
That’s leading IIHS to suggest that automakers rethink how they design the very SUVs American car buyers are snapping up in droves.
“SUVs have higher front ends, and often the design for the vehicle is much more vertical than passenger cars,” IIHS President David Harkey
and headlights that break away on impact, among other ideas. Automakers have mostly focused on pedestrian-detection systems that trigger automatic emergency braking, such as
‘s EyeSight system, which relies on two cameras mounted to the interior roof behind the windshield. IIHS is also urging automakers to develop better headlights, an area it began evaluating as part of its comprehensive crash-test program in 2016.
IIHS is also recommending that officials lower speed limits on busy roads that attract pedestrians and make broader use of
to enforce existing limits. And it says urban planners should design roads with better accommodations for pedestrians, noting that arterial roads often lack mid-road crossing lanes and instead require pedestrians to walk long distances to reach an intersection. Cities and suburbs should also include features like pedestrian-activated beacons or median crossing islands for mid-block crossings, it said. In Detroit, which had the highest pedestrian death rate of major U.S. cities in 2016, the city saw improvements in pedestrian fatalities after installing 65,000 streetlights, according to the
Free Press
.
The 5,987 pedestrian fatalities in 2016, the latest year for which data are available from the
, represented 16 percent of all crash fatalities. Overall, the number of annual pedestrian fatalities is still down 20 percent compared to 1975, but the 2016 toll was the highest since 1990.
Your Smartphone Could Decide Whether You’ll Get a Loan
https://ift.tt/2ryOAbl
Every time you visit a website, you leave behind a trail of information, including seemingly innocuous data, like whether you use an Android or Apple device. And while that might feel like a mere personal preference, it turns out that lenders can use that type of passive signal to help predict whether you’ll default. In fact, new research suggests that those signals can predict consumer behavior as accurately as traditional credit scores. That could disrupt the traditional credit bureau industry that’s dominated since the 1980s—and have serious ramifications for privacy.
In a new working paper from The National Bureau of Economic Research, a team of researchers analyzed over 270,000 purchases from October 2015 to December 2016 on a German e-commerce website that allows customers to buy furniture and pay for it later. (Think of it as Germany’s version of Wayfair.) The store was of particular interest because it already uses a digital footprint, in conjunction with a user’s German credit score, to decide whether buyers qualify for a loan. At least a handful of European retailers have been using similar systems for several years.
The use of a largely outdated email service—like Hotmail or Yahoo—was also an indicator of a higher default rate.
The researchers looked at 10 different types of information customers passively provide, including things like what type of device they used, their operating system, how they got to the site (like whether they clicked on an ad), the time of day they made the purchase, and what kind of email provider they use. The researchers didn’t take into account some factors the retailer normally does, like whether the person has paid a loan back from the same company in the past. Still, they found that those simple variables could be used to estimate whether someone might default, just like a FICO score does.
The difference in default rates between iOS and Android users, for instance, was equivalent to the difference between a median FICO score and the 80th percentile of FICO scores. On one level, these types of insights are intuitive: The average iPhone is much more expensive than the average Android device, and previous research has shown whether someone owns an iOS device is one of the best predictors of whether they’re in the top 25 percent of earners.
The study’s other findings, though, are more subtle. For example, customers who placed orders through cell phones rather than desktop computers were also more likely to default. The use of a largely outdated email service—like Hotmail or Yahoo—was also an indicator of a higher default rate. Customers who incorrectly entered their email address defaulted 5.09 percent of the time; those who didn’t were at .94 percent. In this case, “defaulting” means the loan was sold to a collections agency, usually several months after the purchase and after the customer had been notified three times about their outstanding bill.
Even how you arrive at an e-commerce website can be used to predict whether you’ll default. Those coming in from a price-comparison website were half as likely to default as those who clicked on a targeted ad. That makes sense; savvy, careful consumers browse different retailers’ prices before making a purchase. But even seemingly irrelevant information can say more about your spending behavior than you might expect. For example, customers who have their first or last names in their email addresses were 30 percent less likely to default than those who used something like “cutie367.”
Following Footprints
The researchers ultimately found that digital footprints equaled or exceeded the predictive power of traditional FICO-like credit scores, and could even be used to predict how a person’s FICO score might change in future. The authors say digital data could also potentially be used to assess customers outside of the traditional banking system, who often don’t have FICO scores.
But they also acknowledge that wide use of digital footprints for creditworthiness would likely have serious implications for user behavior and freedom online. Imagine buying an iPhone to qualify for a mortgage, or thinking about car loans when signing up for an email account. Customers fudging their digital footprints could also cause lenders to issue loans to customers that can’t actually pay them back.
“My personal opinion is that among most people, if you have someone who thinks about these types of issues, you’re already talking about people who are financially quite sophisticated,” says Tobias Berg, the study’s lead author and an associate professor at Frankfurt School of Finance & Management. He also points out that most consumers in Germany aren’t aware that information like what type of device they use is sometimes factored into loan approvals, even though it’s explained in retailers’ terms of service agreements. “Almost no one reads that, and no one really understands what it literally means,” he says.
Another concern is that digital footprints might serve as proxies for variables lenders are prohibited by law from taking into account, like race. There are clearly people who “are going to be disadvantaged by these digital footprints, no doubt about it,” says Berg. That includes individuals inadvertently categorized as risky, even when they’re not; plenty of people can afford iPhones but go with Android instead. And there are perfectly valid privacy reasons to leave your name out of your email address, for example.
Berg and his coauthors found some correlation between FICO scores and digital footprints but not much; someone’s FICO score might indicate that they’re qualified for a loan, while their digital footprint says otherwise. Berg accounts for the difference by pointing out that credit scores are fairly crude, and only account for extreme situations like when a customer misses a payment. Digital footprints can reveal more psychologically oriented traits, like how someone thinks about making a purchase or what time of the day they shop.
That’s why the researchers suggest the strongest signal comes from combining the two. But for an unbanked person with only a digital footprint, that disparity might result in being denied a loan they would otherwise get.
Not on the Horizon
The good news is that in the United States, digital-footprint loans are likely a long ways off, in part because companies have found in the past that online information may not be as useful as it seems. “We’ve heard this before. The last iteration was social media; companies saying that they’re going to use your Facebook posts to judge how creditworthy you are,” says Liz Weston, a columnist at NerdWallet and the author of five books, including Your Credit Score. “This stuff sounds scary, but a lot of things don’t affect your credit score now and they’re not likely to in the future.”
That’s partly because the lending industry moves incredibly slowly, and is reluctant to change its methods. “The basic scoring formula has worked pretty well and continues to work pretty well,” says Weston. “I just can’t see it being displaced, and certainly not overnight.” That’s not to say it works perfectly; Weston notes FICO puts minority groups that depend more on cash or informal lending at a disadvantage.
Digital footprints, meanwhile, do come into play somewhat in the US; online retailers have used some of that info to manipulate prices for years.
For now, the general behaviors you need to create good credit aren’t based on whims, like whether you kept your goofy email address from high school. But you can’t predict how websites will analyze and use passive data in the future, especially given how hard it is to avoid disclosing information like what kind of phone you have to a retailer. At the very least, though, you can understand how that information is analyzed—and what conclusions companies draw from it.
