Demystifying the black box of Big Data: Interview with Telefónica R&D’s Angela Shen-Hsieh

Wednesday, 28 June 2017

Demystifying the black box of Big Data: Interview with Telefónica R&D’s Angela Shen-Hsieh

AI and Big Data

Angela Shen-Hsieh is the Director of Predicting Human Behavior in Telefónica Product Innovation. In this role, she is in charge of leading internal innovation. Prior to joining Telefónica in November, Angela worked at IBM Watson, managing the conversational intelligence and data discovery product lines, and was a founder and CEO of several data visualization startups. Angela was trained as an architect, receiving her Master’s in Architecture from Harvard University.


We sat down with Angela to talk about her role at the intersection of Artifical Intelligence, Machine Learning and Big Data.


How did you get your start with technologies like AI, Machine Learning and Big Data, especially given your architecture background?


I actually got involved through design. I ran several companies that designed user interfaces for data, and this turned into a competency around demystifying the black box of analytics so that decision-makers could actually act on the data with confidence. One of the problems is that analytics can be mysterious, so we did a lot of work on internal decision support systems. For example, if you put a number up on a dashboard or give it to a decision maker and they can’t unpack it, they can’t have confidence. They can’t look at it and understand all the implications of it. Say, for example, a given number is red. Why is it red? How long has it been red? Will it still be red if we constrain the data by excluding this one project, one country, or this one product line?

That is a key interactive visualization problem, one that is necessary to really understanding how people use analytics. That’s where the design aspect comes in. Architects, in particular, don’t build buildings, we actually make pictures of buildings. So architects are experts in visual representation. That’s the same approach we took to making data more legible for business people. That turned into building apps, being able to bring analytics to consumers to help them improve their health, improve other types of behaviors and things like that.

AI and Big Data
Figure 1: AI and Machine Learning are key tools for Big Data.

What are some of the biggest obstacles you encounter in using AI and Machine Learning to innovate with Big Data?


The biggest one is the Big Data itself. The term Big Data has been around for a long time, but that does not mean that Big Data is actually as accessible as you would think it would be. There are still the same kind of business intelligence challenges that have been around for forever: data that’s not standardized, that’s not harmonized, that’s not clean, that’s not complete. It’s forensics to figure out and that’s very manual. So Machine Learning and AI offer this automation, but underneath, there is still a lot of manual work. That is one of the challenges.

Another challenge is, as a product organization, we need to figure out how we monetize Big Data and these kinds of technologies. It’s one thing to build a bespoke solution, it’s another problem entirely to commercialize a product that can be sold to many customers at volume and scale: instead of 50 clients each paying €100,000, how do we build a business with 500,000 customers each spending €100 per month? There are all kinds of challenges, beyond solving just the technical machine learning problem of whether the data can tell the story you hope for. At IBM Watson, I saw this up close. We were trying to take this incredible technology that won the TV question-and-answer game show, Jeopardy!, and build a huge business answering customer service questions or finding needles-in-the-haystack of medical literature and more. There were a lot of false starts, a lot of challenges around user expectations, around focus, a lot of pressure to make money doing all of this, and all while the market was moving to commodify these types of natural language query capabilities. To commercialize data and AI/ML—you need to be solving a broader problem.


What brought you to Telefónica?


I came to Telefónica for two reasons: the data and the innovation process. While IBM has incredible breadth and resources as a company of 450,000 people, there are things that Telefónica has that IBM does not. One thing is data.
Fundamentally, these kinds of technologies require a steady diet of data to survive and to improve. I was really intrigued by the possibilities and the breadth and richness of the data Telefónica has available. 
The second thing is the innovation process here at Telefónica. I was really impressed at how thoughtful it is—structured but with a lot of room to experiment and bring in new ideas and people. Not all big companies have such a thoughtful and it takes a very special kind of approach and different kinds of people to take things that may be research assets and turn them into commercial products. I was impressed with that process and that is the area where I particularly like to work: in innovation, which I define as being between more academic research and product development.

Tell us more about your role as Director of Predicting Human Behavior here at Telefónica. What are some current projects that particularly excite you?


There are many things that I am excited about! This area had started out pretty broad, calling for ideas from all over the organization — which is the right approach. This is a moment where now, with the fourth platform that we can more easily build products on top of, where we actually need to bring the projects into a common focus. Still keeping the good things about the innovation process that we have, but getting them more focused to be able to make a bigger impact. The focus that we have in Predicting Human Behavior is what we’re calling Cognitive Customer Experience: how we bring contextual data to improve interactions with customers across the whole customer engagement lifecycle. This may be things that are internal or external, all with an eye towards what is the next business market that Telefónica can capitalize on around data. What’s the future of advertising as we know it, intelligent agents, bots, how purchases and transactions are made, how customer service is carried out?
All of these things are going to be made better, and the intelligence part of artificial intelligence will come through a better understanding of the customer and their context, and that comes through data. So we’re focusing on enabling that.

How do you see AI and Machine Learning changing the way we use Big Data?


In simplistic terms, most traditional analytics have been past-looking. They look backwards and try to tell you things about the past that maybe you would be able to apply to the future. Because these technologies learn over time and they can make correlations and understandings that really aren’t so visible, we should be able to react faster, be more proactive, predict more into the future, and we could then be more prescriptive. As someone who spent decades building dashboards that were mostly reactive, that is an important change evolution.

The other thing is that there is less of a line, when we talk about AI and Machine Learning, between structured and unstructured data — and that’s really important. The industry has long been trying to figure out how to bring those two things together, but they come from very different technological underpinnings. Structured and unstructured data don’t play well together in traditional business intelligence or enterprise search methods. Content and data don’t talk to each other. Yet for the end user, there is little difference between an answer to a question like, “What’s Harrison Ford’s wife’s name?” and “How tall is Harrison Ford?” They think of that as information or data, and, in fact those words are used interchangeably by most people. But in terms of technologies, we store those answers very differently (in a database vs. a content management system) and we go after answers depending on how we store the data (query vs. search.) But using AI and Machine Learning together with Big Data has the promise to bring together structured and unstructured data.

structured and unstructured data
Figure 2: The prospect of bringing together structured and unstructured data can greatly enhance data capabilities. 

What are the biggest areas of untapped potential for Big Data?


In Machine Learning and AI, there are many emerging techniques such as deep learning and neural networks. On the business side, I believe there is going to be syndication of data through a marketplace or network system. Either that or we let Google and Amazon take over the world! In terms of use cases and tangible things that we will feel, I am most interested in how things can help people. We are in a battle in the attention economy, so that would mean making things more friction-free, making things more personal and relevant, and looking at how things can be done with less intervention and less manual connecting of the dots. Right now, we as consumers still have to connect a lot of dots to get end-to-end through any transaction. How can that have less friction and be made safer and easier?

I am also interested in the kinds of use cases where Big Data can actually help me improve myself in ways that I might otherwise struggle to. Things like eating better, getting the right exercise. We’re looking into other areas like applying this to online behavior and mobile phone use behavior. People talk about how disappointed they are in themselves that they cannot pull themselves away from SnapChat or Facebook. Tristan Harris, who is best known for his role with design ethics at Google, has created a “movement” called Time Well Spent. He talks about how we are compelled because techniques have been developed to addict us, and once one app uses that technique, every other app has to follow because they are all competing for our scarce attention. So perhaps we can use Big Data to flip that around. If we can compel and predict our human behavior, we should be able to give that data back to help us actually improve our behavior and other aspects of our lives and ourselves.


No comments:

Post a Comment