Essential Guide

Special Report: Artificial intelligence apps come of age

A comprehensive collection of articles, videos and more, hand-picked by our editors
News Stay informed about the latest enterprise technology news and product updates.

Deep machine learning drives Loop AI quest

Loop AI Labs' Bart Peintner discusses the transformational impact of deep learning technology, artificial intelligence, and how these tech trends will reshape industries.

Bart Peintner has been closely involved with important developments in artificial intelligence through its recent resurgence. At SRI, one of the world's hotbeds of AI research, he pursued work that pressed the limits of natural language processing and user-behavior modeling. Now, as CTO and co-founder of startup Loop AI Labs, he is furthering the cause of unsupervised machine intelligence -- also known as deep machine learning -- for applications. It's important because teaching machines to do human's work can be labor intensive.

When did you start Loop AI Labs, and what was the underlying goal?

Bart Peintner: We founded the company in 2012. At that time we were doing a type of natural language understanding that was focused on understanding people from what they say. As we went along, we found the trouble with then current modeling techniques, and we started experimenting with deep learning. We had a lot of success very quickly, so we kind of changed our focus from just understanding people to understanding text for particular companies or particular industries.

The key attribute of what we did was that we could create models of language and concepts in an unsupervised way. Instead of having people manually tag data, or manually define ontologies, we feed information in, and the software kind of slowly teases apart the meanings of words and builds an ontology on top of that. That is useful, especially for large companies that are just sitting on troves of text data.

Bart Peintner, CTO and co-founder, Loop AI LabsBart Peintner

A lot of people are trying to understand whether cognitive computing is just this year's term for artificial intelligence. What's your take?

Peintner: I see a distinction between cognitive computing and AI, although I understand people have different opinions. For me cognitive computing is focused on either mimicking thought processes for situations where that's important. For example, understanding customer support statements is about analyzing the meanings of a user, their use of language and their intents -- whereas AI, which is where I spent the majority of my career, really doesn't care how you get there. It is about creating a model that helps you make intelligent, accurate predictions. In my mind, it's not necessarily focused on people or how people think and act.

I understand the Loop AI platform uses what's called "deep machine learning." For those among us just getting used to machine learning, could you distinguish what you may mean by deep machine learning?

Peintner: Machine learning underlies a lot of artificial intelligence and a lot of cognitive computing. Our particular type of learning is deep learning applied to text. That's not the traditional type of machine learning that we all know and love.

Deep learning is fairly new, and pretty distinct from what we could call classic machine learning. What makes it distinct is its ability to uncover features in a data set automatically. If you look at classic supervised machine learning, there was a great amount of art required in defining the features of a data set. After that was created, the machine learning software could build models over those data sets. But the value of deep learning is that it uncovers these features automatically, and that is how it has been able -- for speech recognition, as an example -- to surpass decades of research in just a few years. 

Can you talk about the use cases where you have found this is helpful?

Peintner: One real sweet spot for this technology is customer support. What makes customer support hard for the older techniques is that every company has its own products, its own concepts and its own problems with products. So to model all of those nuances takes a lot of people who are either studying the domain or building the model, or it takes a lot of hand-labeling of customer support tickets. So it is helpful to have a tool into which you can feed a bunch of language. If it can pull out the meanings of words in that context, it makes it possible in an affordable way to model one company's data very specifically.

You have been at AI for a long time. What is your perspective on recent deep machine learning developments?

Peintner: When the deep learning results started surfacing, I -- like many others who have been around for a while -- was very skeptical. It wasn't until I saw results on problems I was running that I understood its transformational impact. But it is not transformational in the way you sometimes see in the press -- which talks about killer robots and sentient beings, and which you will not see anytime soon -- but it is transformational nonetheless.

What is happening, and very quickly, is that deep learning is able to produce models that enable knowledge work to be done very efficiently. And, just as self-driving cars are going to start replacing some jobs, deep learning applied to text and images and videos is going to start displacing knowledge workers' jobs as well. It really is surprising the capabilities this technology has to uncover patterns and features.

Here's a question on markets. IBM's Watson seems like a dual-edged sword for the cognitive market. Some people know about Watson but nothing else. What's your take on that?

Peintner: Well, IBM is educating the enterprise community of the power and potential of cognitive computing. We are benefiting from the increased awareness and sense of urgency that IBM has created around cognitive computing. But, at the end of the day, IBM Watson and Loop AI are competitors, as we're both enabling cognitive applications for large enterprises. Loop AI's platform's ability to learn and reason on its own without requiring human supervision competes directly with IBM Watson's enterprise offering, which requires a long time to build a cognitive application using a mix of software and on-site human expertise.

Next Steps

See how machine learning is being applied in IoT

Read about trends in hiring for AI skill sets

Go back to the roots of AI with this homage to one of the greats

How AI and deep learning are steering domain-driven design

This was last published in March 2016



Find more PRO+ content and other member only offers, here.

Essential Guide

Special Report: Artificial intelligence apps come of age

Join the conversation


Send me notifications when other members comment.

By submitting you agree to receive email from TechTarget and its partners. If you reside outside of the United States, you consent to having your personal data transferred to and processed in the United States. Privacy

Please create a username to comment.

What types of applications could your organization apply deep learning to?
When we are considering identifying patterns in multimedia, specifically mixes of images and video with or without associated text that can be linked to individuals and segments of audiences publicly available in social media, new paradigms of understanding can be created around what is being said and visualized, as well as what is missing. 

It will be interesting to see how deep machine learning is applied, which specific industries are willing to gamble first in applying to to their needs, and how smart these organizations will be in deploying across their entities or if the technology gets siloed with the IT/data teams.
Customer experience, what they had problems with, and what they liked. Companies spend fortunes trying to predict the acceptance and miss on the actual feedback.