In the past, artificial intelligence (AI) has been unable to measure up to the lofty expectations set by the thinking machines in the movies, such as HAL in 2001: A Space Odyssey. Instead, the technology repeatedly climbed the “hype cycle” only to fall into a “trough of despair.” This is Part 1 of an interview with Monty Barlow, director of machine learning at Cambridge Consultants, and he reveals why this time is different—and why there’s no going back.

How has AI evolved?

At Cambridge Consultants, we’ve been doing AI long enough to have lived through several cycles of hype and disappointment. The 1990s were a time of great excitement around AI. At that time, AI meant decision support, fuzzy logic, and expert systems. Then AI entered one of its winters, people stopped using the term, and companies lost interest. We were still creating intelligent systems for all sorts of things, but we weren’t calling them AI.


Sentiment started to shift around 2006 with the release of new GPU technologies, the growth of internet data, and the introduction of new algorithms. Today, deep learning and the neural networks that enable it are expanding at a breakneck pace. Each advancement is laying a foundation for making the unimaginable real.


We’ve never before seen a technology that’s so relevant in every single market. With deep learning in particular, AI began advancing so quickly that we knew we had to do our own independent research in order to help our customers create disruptive solutions. That’s when we decided to build our own research lab dedicated to discovering, developing, and testing machine learning approaches. We call it the Digital Greenhouse.

How are you helping clients innovate with AI?

One example is in industrial applications where old-school machine vision is being replaced by smarter, deep learning. Automobile manufacturers are using it in their development of autonomous vehicles. Healthcare providers use AI to improve cancer screening and provide better preventive care. Manufacturing companies use AI for machine maintenance and QA testing. Energy companies use it for resource exploration and efficiency. And the list goes on.


The work we do with our clients is highly confidential. We cannot discuss the details because they are deploying AI to gain a competitive edge. That’s why we created the VincentTM AI artist.

What is the Vincent AI system and how does it work?

To showcase the possibilities of deep learning, we developed an interactive system called Vincent. We started with what we thought was an impossible task—have a machine turn a simple sketch into a work of art that could have been created by a master. With Vincent, people of all ages and backgrounds get to play with technology that confounds their expectations—technology that was impossible to imagine only a few years ago.


Vincent learned from exposure to thousands of masterpieces from Van Gogh, Cézanne, Picasso, and many others. During training, multiple generative adversarial networks competed to tune the algorithm so that Vincent can produce credible works of art by drawing inspiration from these masters.


A generative adversarial network is a technology in which neural networks challenge each other during training to get better at synthesizing data and spotting data that’s synthesized versus real data. By the way, Vincent was trained entirely by using NVIDIA GPUs and NetApp® storage.


Getting a machine to turn a few lines into a seamless, colorful visual is easier said than done. We picked a particularly difficult problem, thinking that if we could teach a machine to translate from simple lines to a painting, imagine what else we could teach it to do.


Why should people take AI seriously now?

Earlier promises of AI didn’t pan out for good reasons. Datasets were small and computers lacked power, thwarting experiments that were otherwise on the right tracks. Even the mighty imagination of the early AI pioneers could not drive the field forward against this constant failure.


Today, with advances in technology and access to vast datasets, machines are learning by themselves, giving them the ability to outthink and outperform humans.


With neural networks and deep learning, AI can exceed our wildest imaginations. We’re at the beginning of a new era—fueled by data and enabled by new technologies—and we are writing the story as we go.


Watch this short video to see the Vincent AI system in action. Part 2 of the interview will cover the technology used to train Vincent.


Turning our sketches into art with machine learning from Cambridge Consultants on Vimeo.

Monty Barlow

Monty Barlow is director of machine learning for Cambridge Consultants and an expert in the on-time and on-spec development of technology deemed impossible by others. Cambridge Consultants develops breakthrough products, creates and licenses intellectual property, and provides business consultancy in technology-critical issues for clients worldwide.

Add comment