All posts in “AI”

Microsoft Pix can scan business cards to your contacts, find people on LinkedIn

LinkedIn used to have its own business card scanning app, CardMunch, which served a useful purpose in a world where paper cards simply refuse to die. But that app was shut down back in 2014, with LinkedIn suggesting users move to Evernote instead. Today, Microsoft is bringing back business card scanning – but this time, not with a dedicated card scanner app, but with its multipurpose, A.I.-powered camera app, Microsoft Pix.

Since its launch in 2016 as an iOS app that helps you take better pictures, Microsoft has increasingly found more productivity-related uses for Pix. In September, for example, the app was updated to include a way to snap better photos of documents, post-its, whiteboards, and yes, business cards.

But with today’s update, Pix’s business cards smarts are being upgraded – this time with a LinkedIn integration. In the latest version of the iOS app, Pix includes a new business card feature that will add new contacts both to your iPhone’s address book, as well as to your LinkedIn account.

To take advantage of this option, you just launch the app and point it at the business card. Pix then automatically detects what it’s seeing, and asks you if you want to “Add Contact” or “Find on LinkedIn.”

When you tap to add the contact, Pix captures and organizes the contact information – like name, phone, address, and URL – into the correct fields, and adds the newly created contact to your iPhone’s Contacts app. If you opt for LinkedIn, you’re able to view the person’s profile in the LinkedIn app on your iPhone, and optionally add them to your list of connections.

The business card scanning feature, like others in Pix, leverages A.I. technology under the hood to enhance and improve the image. In the case of business cards, Pix is able to detect the edges of the cards, sharpen focus, and tweak the angle of the photo to render the image in a straight-on perspective so it can extract the information from the card.

The Pix update is just one of several ways Microsoft has integrated with LinkedIn since acquiring the company for $26.2 billion in 2016. It has also tied LinkedIn into its other products, including Office 365,, Dynamics 365, Word, and Windows 10.

The updated version of Microsoft Pix is rolling out today. You may not have it yet, as it has to propagate across the App Store, so keep your eyes peeled.

ELSA raises $3.2M for its A.I.-powered English pronunciation assistant

ELSA, an app whose name stands for “English Language Speech Assistant” (and not the popular Disney character!), has raised $3.2 million for its A.I.-assisted language learning platform that teaches people how to speak English. Unlike other courses that focus mainly on teaching grammar and vocabulary, ELSA uses artificial intelligence and speech recognition technology to help language learners with their pronunciation.

The $3.2 million pre-A round of funding was led by Monk’s Hill Ventures, a firm that invests in post-seed stage startups in Southeast Asia. Monk’s Hill founder and partner, Peng T. Ong, is joining ELSA’s board.

The San Francisco-based startup was originally founded in 2015 by Stanford grad Vu Van, ELSA CEO, and Dr. Xavier Anguera, whose background is in speech recognition and A.I. technologies. It debuted at SXSW in March 2016, where it later won the SXSWedu launch competition.

According to Van, who was born and raised in Vietnam, the idea for ELSA was prompted by her personal experience in trying to learn English.

“I moved to the States for my MBA and Master’s in Education at Stanford,” she says. “My first year at Stanford was very challenging because of my ability to speak English. A lot of the time, people misunderstood me,” Van continues.

Although ELSA’s founder was able to write and read English fairly well, she wanted to find a good solution for improving her accent – and she came up short.

“I realized that people didn’t really have a lot of solutions…when it comes to speaking, they could either go to a speech therapist that costs them $150 an hour, who could listen to them and fix their pronunciation, or they could go to YouTube or watch Netflix, which is a one-way learning solution,” Van says.

She decided to develop ELSA as a result of her own struggles, bringing in co-founder Dr. Anguera to help create ELSA’s proprietary speech recognition technology.

To use ELSA, language learners download the app for iOS or Android, then take ELSA’s five-minute assessment test which identifies the user’s pronunciation proficiency and identifies where they still have challenges. This information is then used to build out a personalized curriculum, tailored to the user’s current abilities.

In ELSA, there are around 600 lessons and 3,000+ words across a variety of topics, like introductions, talking about family or relationships, jobs and hiring, traveling, and more. The app is also updated regularly with seasonal and timely topics, like lessons featuring the holidays, or even the new “Star Wars” movie, so learners can better participate in everyday communication.

The lessons themselves are bite-sized, at 2 minutes long. They have five exercises that get progressively harder, including pronunciations of words, phrases and sentences.

ELSA works by listening to learners’ voices, then matching up what they said with the correct American English pronunciation. The words on the screen are highlighted in red, yellow and green to show how well each the student did. The app will also help by making suggestions as to how the speaker can improve on a given sound – for example, by telling them how to shape their mouth or move their tongue.

Behind the scenes, ELSA is powered by A.I. technology, that listens to users’ speech – something the team built from scratch.

“The speech recognition technology out there is slightly different [from ELSA]. They try to guess what you’re saying, whether you’re saying it right or wrong. It’s obviously very forgiving to mistakes. What we want to do is the exact opposite,” says Van.

Since its launch two years ago, ELSA has been adopted by a few million users in over 100 countries, with around half its user base in Southeast Asia, and the rest spread out elsewhere in the world, including in Latin America and Eastern Europe. Users are today practicing a few million exercises per week.

With the additional funding, ELSA is planning to address the requests from teachers who want to use the service in their classrooms by developing a tool that will let them enter words and phrases from own curriculum. It’s also going to begin the work to add new languages to the platform, and build out its voice recognition API.

ELSA today generates revenue through subscriptions, priced at $3.99/month or $29.99/year. A quarterly option is available as well, though prices may vary by region. While the product for teachers will be affordable, it won’t be free – potentially offering another revenue stream.

The company will also use the funds to hire additional A.I. talent, says Vu.

ELSA’s app is a free download on iOS and Android.

SafeToNet demos anti-sexting child safety tool

With rising concern over social media’s ‘toxic‘ content problem, and mainstream consumer trust apparently on the slide, there’s growing pressure on parents to keep children from being overexposed to the Internet’s dark sides. Yet pulling the plug on social media isn’t exactly an option.

UK startup SafeToNet reckons it can help, with a forthcoming system of AI-powered cyber safety mobile control tools.

Here at Mobile World Congress it’s previewing an anti-sexting feature that will be part of the full subscription service — launching this April, starting in the UK.

It’s been developing its cyber safety system since 2016, and ran beta testing with around 5,000 users last year. The goal is to be “protecting” six million children by the end of this year, says CEO Richard Pursey — including via pursuing partnerships with carriers (which in turn explains its presence at MWC).

SafeToNet has raised just under £9 million from undisclosed private investors at this point, to fund the development of its behavioral monitoring platform.

From May, the plan is to expand availability to English speaking nations around the world. They’re also working on German, Spanish, Catalan and Danish versions for launch in Q2.

So what’s at stake for parents? Pursey points to a recent case in Denmark as illustrative of the risks when teens are left freely using social sharing apps.

In that instance more than 1,000 young adults, many of them teenagers themselves, were charged with distributing child pornography after digitally sharing a video of two 15-year-olds having sex.

The video was shared on Facebook Messenger and the social media giant alerted US authorities — which in turn contacted police in Denmark. And while the age of consent is 15 in Denmark, distributing images of anyone under 18 is a criminal offense. Ergo sexting can get even consenting teens into legal hot water.

And sexting is just one of the online risks and issues parents now need to consider, argues Pursey, pointing to other concerns such as cyber bullying or violent content. Parents may also worry about their children being targeted by online predators.

“We’re a cyber safety company and the reason why we exist is to safeguard children on, in particular, social networking and messaging apps from all those things that you read about every day: Cyber bullying, abuse, aggression, sextortion, grooming,” he says.

“We come from the basis that existing parental control systems… simply aren’t good enough. They’ve not kept up to date with the digital world and in particular the world that kids socialize on. So Snapchat, Instagram, less so Facebook, but you get the idea.

“We’ve tackled this using a whole mixture of deep tech from behavioral analytics, sentiment analysis and so on, all using machine learning, to be able to contextualize messages that children send, share and receive. And then block anything harmful. That’s the mission.”

Once the SafeToNet app is installed on a child’s device, and linked with their parents’ SafeToNet account, the software scans for any inappropriate imagery on their device. If it finds anything it will quarantine it and blur the content so it no longer presents a sharing risk, says Pursey.

The software runs continuously in the background on the device so it can also step in in real-time to, for instance, block access to a phone’s camera if it believes the child might be about to use it for sexting.

It’s able to be so reactive because it’s performing ongoing sentiment analysis of everything being typed on the device via its own keyboard app — and using its visibility into what’s being sent and received, how and by whom, to infer a child might be about to send or see something inappropriate.

Pursey says the AI system is designed to learn the child’s normal device usage patterns so it can also alert parents to potential behavioral shifts signaled by their online activity — which in turn might represent a problem or a risk like depression or aggression.

He says SafeToNet’s system is drawing on research into social behavioral patterns, including around digital cues like the speed and length of reply, to try to infer psychological impacts.

If that sounds a little Black Mirror/Big Brother, that’s kind of intentional. Pursey says it’s deliberately utilizing the fact that the children who are its users will know its system is monitoring their device to act as a moderating impulse and rein in risky behaviors.

Its website specifies that children have to agree to the software being installed, and kids will obviously be aware it’s there when it pops up the first notification related to something problematic that they’re trying to do.

“If children know they’re being watched they automatically adjust their behavior,” he says. “We’re using a high degree different methods to deploy our software but it is based upon research working with universities, child welfare support groups, even a priest we’ve been talking to.”

On the parent side, the system hands them various controls, such as enabling them to block access to certain apps or groups of apps for a certain time period, or lock out their kids’ devices so they can’t be used at bedtime or during homework hours. Or ground access to a device entirely for a while.

Though, again, SafeToNet’s website suggests parents use such measures sparingly to avoid the tool being used to punish or exclude kids from socializing digitally with their friends.

The system can also report on particular apps a child is using that parents might not even know could present a concern, says Pursey, because it’s tracking teen app usage and keeping an eye on fast-changing trends — be it a risky meme or something worse.

But he also claims the system is designed to respect a child’s privacy, and Pursey says the software will not share any of the child’s content with their parents without the child’s say so. (Or, in extremis, after a number of warnings have been ignored by the child.)

That’s also how he says it’s getting around the inevitable problem of no automated software system being able to be an entirely perfect content monitoring guardian.

If/when the system generates a false positive — i.e. the software blocks content or apps it really shouldn’t be blocking — he says kids can send a request to their parents to unlock, for example, an image that wasn’t actually inappropriate, and their parents can then approve access to it.

Another privacy consideration: He says SafeToNet’s monitoring systems are designed to run without any of its employees accessing or viewing children’s content. (You can read the company’s Privacy Policy here. They’ve also written a plain English version. And published a privacy impact assessment.)

Though the vast majority (circa 80%) of the data processing it needs to do to run this pervasive monitoring system is being done in the cloud right now. So it obviously cannot guarantee its systems and the data being processed there are safe from hacking risks.

Asked about the company’s intentions towards the user data it’s collecting, Pursey says SafeToNet will not be selling usage data in any form whatsoever. Activity data collected from users will only be used for making improvements to the SafeToNet service itself, he emphasizes.

But isn’t deploying background surveillance of children’s digital devices something of a sledgehammer to crack a nut approach to online safety risks?

Shouldn’t parents really be engaging in ongoing and open conversations with their children in order to equip them with the information and critical thinking for them to be able to assess Internet risks and make these kind of judgement calls themselves?

Pursey argues that risks around online content can now be so acute, and kids’ digital worlds so alien to parents, that they really do need support tools to help them navigate this challenge.

SafeToNet’s website is also replete with warnings that parents should not simply tune out once they have the system installed.

“When you realize that the teenage suicide rate is through the roof, depression, all of these issues you read about every day… I don’t think I would use that phrase,” he says. “This isn’t about restricting children it’s actually about enabling their access to social media.

“The way we look at is the Internet is an incredibly powerful and wonderful thing. The problem is is that it’s unregulated, it’s out of control. It’s a social experiment that nobody on the planet knows how it’s going to come out the other end.”

“I’ve seen a 10 year old girl hang herself in a cupboard,” he adds. “I’ve seen it. I saw it online. I’ve seen two 12 year old boys hang themselves. This morning I saw a film of two Russian girls jumped off a balcony to their death.

“I’ve seen a man shot in the head. I’ve seen a man — two men, actually — have their heads chopped off. These are all things that six year old kids can stumble across online. When you’ve seen those sorts of things you can’t help be affected by them.”

What about the fact that, as he says, surveillance impacts how people behave? Isn’t there a risk of this kind of pervasive monitoring ending up constraining children’s sense of freedom to experiment and explore boundaries, at a crucial moment when they are in the process of forming their identities?

A child may also be thinking about their own sexuality and wanting private access to information to help them try to understand their feelings — without necessarily wanting to signpost all that to their parents. A system that’s monitoring what they’re looking at and intervening in a way that shuts down exploration could risk blocking natural curiosity and even generate feelings of isolation and worse.

“Children are trying to determine their identity, they’re trying to work out who they are but… we’re not there to be the parent,” Pursey responds on that. “We’re they’re to advise, to do the safeguarding… But [parents’ job] is to try and make sure that their children are well balanced and well informed, and can handle the challenges that life brings.

“Our job is certainly not to police them — quite the opposite. It’s to enable them, to give them the freedom to do these things. Rather than sledgehammer to crack a nut, which is the existing parental control systems. In my opinion they cause more harm than they actually save or protect. Because parents don’t know how to use them.”

SafeToNet’s software will work across both Android and iOS devices (although Pursey says it was a lot easier to get it all working on Android, given the open nature of the platform vs Apple’s more locked down approach). Pricing for the subscription will be £4.99 monthly per family (with no limit on the number of devices), or £50 if paid up front for a year.

Featured Image: panco971/Shutterstock

Google’s new AI can predict heart disease by simply scanning your eyes

Image: ben brain/Digital Camera Magazine via getty images

The secret to identifying certain health conditions may be hidden in our eyes. 

Researchers from Google and its health-tech subsidiary Verily announced on Monday that they have successfully created algorithms to predict whether someone has high blood pressure or is at risk of a heart attack or stroke simply by scanning a person’s eyes, the Washington Post reports

Google’s researchers trained the algorithm with images of scanned retinas from more than 280,000 patients. By reviewing this massive database, Google’s algorithm trained itself to recognize the patterns that designated people as at-risk. 

This algorithm’s success is a sign of exciting developments in healthcare on the horizon. As Google fine-tunes the technology, it could one day help patients quickly and cheaply identify health risks. 

But don’t get too excited yet. The algorithm didn’t outperform existing medical approaches, such as blood tests according to the Washington Post report. The algorithms were able to pick out the patient at risk 70 percent of the time. That’s impressive, but it’s far from perfect. 

The procedure also hasn’t been replicated or validated to the point where it can be broadly accepted in the scientific community. 

And experts don’t think it will be necessary for Google’s technology to replace conventional, human-powered care in the near future. 

Maulik Majmudar, associate director of the Healthcare Transformation Lab at Massachusetts General Hospital, told the Post that age and gender are already good predictors of risk for such disease. While Google’s algorithm is an improvement, its improvement to current healthcare practices would be marginal at best. 

That said, it’s clear that artificial intelligence and machine learning have the potential to bring added convenience and affordability to the healthcare industry, even in areas as small as our eyes. 

[embedded content]

Amazon may be developing AI chips for Alexa

The Information has a report this morning that Amazon is working on building AI chips for the Echo, which would allow Alexa to more quickly parse information and get those answers.

Getting those answers much more quickly to the user, even by a few seconds, might seem like a move that’s not wildly important. But for Amazon, a company that relies on capturing a user’s interest in the absolute critical moment to execute on a sale, it seems important enough to drop that response time as close to zero as possible to cultivate the behavior that Amazon can give you the answer you need immediately — especially, in the future, if it’s a product that you’re likely to buy. Amazon, Google and Apple are at the point where users expect technology that works and works quickly, and are probably not as forgiving as they are to other companies relying on problems like image recognition (like, say, Pinterest).

This kind of hardware on the Echo would probably be geared toward inference, taking inbound information (like speech) and executing a ton of calculations really, really quickly to make sense of the incoming information. Some of these problems are often based on a pretty simple problem stemming from a branch of mathematics called linear algebra, but it does require a very large number of calculations, and a good user experience demands they happen very quickly. The promise of making customized chips that work really well for this is that you could make it faster and less power-hungry, though there are a lot of other problems that might come with it. There are a bunch of startups experimenting with ways to do something with this, though what the final product ends up isn’t entirely clear (pretty much everyone is pre-market at this point).

In fact, this makes a lot of sense simply by connecting the dots of what’s already out there. Apple has designed its own customer GPU for the iPhone, and moving those kinds of speech recognition processes directly onto the phone would help it more quickly parse incoming speech, assuming the models are good and they’re sitting on the device. Complex queries — the kinds of long-as-hell sentences you’d say into the Hound app just for kicks — would definitely still require a connection with the cloud to walk through the entire sentence tree to determine what kinds of information the person actually wants. But even then, as the technology improves and becomes more robust, those queries might be even faster and easier.

The Information’s report also suggests that Amazon may be working on AI chips for AWS, which would be geared toward machine training. While this does make sense in theory, I’m not 100 percent sure this is a move that Amazon would throw its full weight behind. My gut says that the wide array of companies working off AWS don’t need some kind of bleeding-edge machine training hardware, and would be fine training models a few times a week or month and get the results that they need. That could probably be done with a cheaper Nvidia card, and wouldn’t have to deal with solving problems that come with hardware like heat dissipation. That being said, it does make sense to dabble in this space a little bit given the interest from other companies, even if nothing comes out of it.

Amazon declined to comment on the story. In the mean time, this seems like something to keep close tabs on as everyone seems to be trying to own the voice interface for smart devices — either in the home or, in the case of the AirPods, maybe even in your ear. Thanks to advances in speech recognition, voice turned out to actually be a real interface for technology in the way that the industry thought it might always be. It just took a while for us to get here.

There’ a pretty big number of startups experimenting in this space (by startup standards) with the promise of creating a new generation of hardware that can handle AI problems faster and more efficiently while potentially consuming less power — or even less space. Companies like Graphcore and Cerebras Systems are based all around the world, with some nearing billion-dollar valuations. A lot of people in the industry refer to this explosion as Compute 2.0, at least if it plays out the way investors are hoping.