AI Weekly: Hurray for boring AI Posted on : Sep 13 - 2019

I keep having the same conversation with companies that want to brief me on their latest AI-related news. It goes something like this: They have some new feature or improved stats or fresh achievement that’s specific to their business. It’s a great win for them internally, but not exactly the most compelling news in and of itself. Then we get to the AI part, and I start to salivate, asking them to tell me more. The reply is increasingly some version of “We use [a mostly unexciting AI tool] to do it.”

Being accustomed to covering technology for technology’s sake, I always feel a little let down. Given how rich, complex, and world-changing AI technologies can be, I keep expecting a dazzling tale of technological magic.

But I’ve come to see my lukewarm response as immature and arrogant. I was thinking like kids who see their parents — who have great jobs and provide a nice life for them — as severely uncool because they aren’t professional athletes or movie stars. Sure, AI is sexy when it’s theoretical or a moonshot, but in the real world the smartest people in the room are those actually using it to do things for their companies.

Here’s just a smattering of recent examples:

Booking.com needs to predict its customers’ travel needs in volume, from flights to lodging to events. As the service has to be personal and make excellent recommendations, the company’s solution is to build a graph for its customers so their questions can be answered as quickly as possible. This involves quite a mix of AI. Ram Papatla, VP of experiences at Booking.com told me in an interview that there may be as many as 20 AI techniques employed on the Booking.com homepage at any given time. That mix includes chatbots and numerous machine learning models, but these, Papatla pointed out, are just commodities. The key is the customer graph.

ZipRecruiter is using a deep learning-based recommendation algorithm and natural language processing (NLP) to match potential employees and employers. The process started with a large investment in AI, first in machine learning and then in deep learning. The company used “candidate calibration” to train the algorithm on what particular employers were looking for and ideally match them to job candidates. But there was still a disconnect — a social engineering problem — because it was taking too long for employer and employee to make the needed social connection. And ZipRecruiter realized it would be better to have a company recruit job seekers than have individuals apply.

Part of the solution to this problem is Job Seeker Profiles, an AI-powered tool that helps job candidates gussy up their profiles to look better to potential employers. For example, if the system sees that you entered “I am a nurse,” it will automatically ping you to add more information and better clarify what that really means. It also gives job applicants insight into who is looking at their profile and how. View More