(GIF via YouTube)
Last week, NYU Tandon’s Future Labs held its AI Summit, an afternoon session devoted to the latest happenings in artificial intelligence — and what that actually means for businesses looking to use or build it. (See our recap here.) The event was equally a showcase for the first-ever cohort of the university’s AI NexusLab, the incubator that it established jointly with ff Venture Capital last summer. Five companies were chosen from hundreds of applicants from around the world, and at the summit, their CEOs pitched the audience on their latest developments.
(The incubator, by the way, is currently accepting applications for its next program, which begins in July. The deadline to apply is May 3.)
All of the AI NexusLab companies, notably, have a business-to-business focus.
Most of them have homed in on a particular sector of interest: Behold.ai, for instance, addresses healthcare, while Alpha Vertex is tackling financial services. Those two characteristics dovetail nicely with the traits for success that Matt Turck, the managing director of FirstMark Capital, identified later during the event as advantages for AI startups. Right now, he said, many companies are still trying to figure out what makes artificial intelligence practical for them, not just cool, so having specific industry applications will give startups a leg up.
Indeed, that emphasis on applicability was paramount in each of the CEOs’ presentations. Their pitches all featured case studies from partners the companies have worked with to develop their initial products and services. Two of the companies even brought representatives from those partner companies onto the stage: Benedikt Battke of German auto company Daimler, for instance, talked up the smart robotics capabilities of Cambrian Intelligence. The incubator, it seems, is striving to make it clear that its AI ambitions aren’t the pie-in-the-sky imaginings often bandied about in the press; its companies are achieving concrete goals right now.
Now, onto the companies…
A smarter investment manager
— Jennifer Litorja (@jenniferlitorja) April 5, 2017
First up was Mutisya Ndunda, the CEO of Alpha Vertex, whose company has developed AI-powered analysis and forecasting for financial markets. As Ndunda, who previously worked at Bloomberg and Merrill Lynch, acknowledged, investment firms such as BlackRock are gradually turning toward tech to gain an edge in financial performance. But finance is still far from an AI takeover: according to the stats Ndunda referenced, only 1 percent of investment firms are using it.
Alpha Vertex has developed three services to date: PreCog, a forecasting service that tracks 30,000 global stocks; SentryWatch, an investment research service that monitors financial news and Rob Oto, an AI financial expert that draws upon the aforementioned services to perform custom analysis. (Rob uses natural language processing to respond to customers’ inquiries.) With these tools, Ndunda said, Alpha Vertex can create a custom investment model for each of its clients. The company is also working with Quantopian, whose platform enables its users to write investment algorithms and to be compensated based on those algorithms’ performance.
An actually useful customer service bot
— ff Venture Capital (@ffvc) April 5, 2017
James Fan, the CEO and cofounder of HelloVera, has a distinctive claim to fame. Previously at IBM, he led the team that developed the Watson project that defeated Jeopardy! champions Brad Rutter and Ken Jennings. His cofounder Liangliang Cao, an adjunct professor at Columbia, is no slouch either: He and his colleagues won the ImageNet Large Scale Visual Recognition Challenge, a prominent competition in computer vision, in 2010. At first glance, their company’s product seems rather mundane: another customer service chatbot? But as Fan explained, most chatbots right now just aren’t very good.
“With all that money spent, nobody likes customer service,” he said.
Namely, chatbots follow a script, but customers often have questions that aren’t accounted for in that script. Vera, on the other hand, is designed to integrate with a company’s complete knowledge base so that it can address tasks such as order changes and return requests, which often have to be handed off to human employees. HelloVera’s customers include Beachbody, of P90X fame, and Tough Mudder.
A system that brings medical billing into the 21st century
Jeet Raut, the CEO of Behold.ai, opened his presentation with an astonishing stat: nearly half of Medicare claims include errors. As you might imagine, that leads to a lot of wasted time and resources in correcting those claims so that providers get paid. In fact, according to Raut’s numbers, $10.6 billion is spent in the U.S. each year on medical billing and coding. Behold.ai seeks to cut that spending significantly.
One big problem, Raut says, is that even if providers use electronic health records, those records often need to be printed and faxed for billing purposes, then converted back into electronic form. That process leaves plenty of room for error. Behold.ai, by contrast, automatically parses records for necessary billing information and converts it into the proper format for filing claims. Its AI system adapts to feedback: in other words, whether the claims it prepares are accepted or not. It’s a seemingly simple task, but parsing human language for critical information is quite tricky, especially when it comes in such an arcane form as medical records. As Dan Lambert, an advisor to the company, put it, it’s “the self-driving car of medical billing.”
An easy-to-use training system for robots
— Future Labs (@NYUFutureLabs) April 5, 2017
In doomsday scenarios for artificial intelligence, brilliant but evil robots usually feature quite heavily. But look at today’s robots, and the notion is pretty easy to laugh at: even the most “intelligent” ones are often pretty clumsy. In order to carry out the precise tasks required for industrial applications, each of their movements has to be meticulously programmed. What happens when, say, a factory’s assembly process changes with the launch of a new product? Yep, those robots have to be completely reprogrammed.
That’s where Cambrian Intelligence comes in. “Software is the real limitation of us having a fully automated world,” CEO Miika Perä told the audience. The London-based company is developing an AI platform that makes industrial robots much more adaptable. It’s currently working with Daimler, which has launched an initiative devoted to what the company calls “intuitive mobility.” That means not just self-driving cars but artificial intelligence applications throughout the company, including sales and supply chain.
A platform to track customers in the real world
— Future Labs (@NYUFutureLabs) April 5, 2017
As most of us know by now, companies such as Facebook and Google have compiled pretty extensive dossiers on how we all spend our time online. Businesses are looking to do the same offline, through technologies such as geofencing and beacons — all with the goal of serving us more targeted ads and ultimately ushering us into their doors. But according to Klustera CEO Pablo Martinez, those technologies have limited accuracy and reach.
Klustera, based in Mexico City — “We’re not paying for your wall,” Martinez joked during his presentation — uses public WiFi networks and closed-circuit television feeds from businesses to track customers’ movements, thereby gathering a more complete picture of people’s activities. (Don’t worry; this information is anonymized, Martinez told the audience.) As a result, the company can give businesses more nuanced reports on their customers’ affinities, based upon their real-world activities: say, a store’s owner might learn that her customers favor a particular type of restaurant, so she can place ads accordingly. Klustera is already operating in Mexico and is in the process of signing up U.S. customers, including a leading quick-service restaurant chain.