Facebook, Salesforce, Google and IBM Research Offer AI Updates at MIT


By John P. Desmond, AI Trends Editor

Here is a roundup from the MIT Technical Conference held on Feb. 18, with its focus on Exponential Technologies:

Facebook is working on “social” virtual reality, following up on its purchase of Oculus in 2014. Manohar Paluri, research lead at Facebook, a co-founder of Facebook AI Research, who has worked for over 10 years on computer vision research, described efforts to make realistic, 3-D, full-body avatars. “The face and hands are the two most important parts to make progress on,” he said. Also, “Recreating hair in the VR world is a tough problem.”

Salesforce last September introduced Salesforce Einstein, with the tagline “AI for Everyone,” bringing AI capability into the core of the Salesforce CRM platform to help sales, service and marketing. Initial capabilities include machine learning models that operate against Salesforce marketing and sales data to enable predictive forecast, optimization and personalization.

Richard Socher, chief scientist at Salesforce , was the CEO and founder of MetaMind, a startup acquired by Salesforce in April 2016 that had been focused on deep learning from image and text data. He walked through a brief history of AI development. “Speech recognition works a lot better; natural language understanding has also made great progress; computers are faster,” he said; however, “AI is only as good as the training data we give it. So it’s important we have a diverse workforce to work in AI.” In response to a question on whether we are in an AI hype cycle, Socher said, “There are good reasons why there is so much excitement. This stuff works.”

Google is working on quantum computing, computers that process in qubits, which is a quantum analogue of the classical bit. (Quantum can be defined as the smallest possible discrete unit of any physical property, such as energy or matter. The term was first used in 1900 by physicist Max Planck, in a presentation to the German Physical Society on radiation.) Computer scientists have theorized that quantum computer could vastly outperform classical computers at certain tasks. Getting the quantum computer out of the lab has been an elusive quest; some in the industry now believe is in the lead and close to a breakthrough.

At the MIT Technical Conference, John Martinis, lead of the Quantum Computing Group at Google, was lowering expectations that any breakthrough would happen soon. He joined the Google team in 2014, after spending much of his career in pioneering research on superconducting quantum bits, qubits. “I’ve been working on this for 30 years,” he said. “We now kind of know what we’re doing. We’re starting to see the end of Moore’s Law. It’s hard to build a quantum computer. At first, it will probably be academic. Once it works, we will build it to a larger system. It could be a ten-year time frame.”

Also, “You need special purpose algorithms to use quantum computing. We’re working on it. As a hardware guy, I’m always thinking about how to get it to work.” Asked what is the practical application of a quantum computer, Martinis said, “Machine learning, solving quantum chemistry problems,” as examples. “We’re only one small idea away from doing something very useful.” Asked if quantum computing is also in a hype cycle, he said, “There has been some hype in quantum computing, but now it’s working well enough that big companies have to invest in it, to see where it goes.”

IBM Research was demonstrating its TJBot open source project, designed to help developers and students gain access to Watson Services. For now, design files can be downloaded so the user can print the parts on a 3D printer in order to build the TJBot, which is a little box-shaped robot. The step-by-step instructions used to help connect TJBot to Watson services are called Recipes. The TJBot maker kit originated from IBM Research in Yorktown Heights, NY, as an experiment to find the best practices in the design and implementation of cognitive objects.

Initial Recipes allow TJBot to respond to emotions, to voice control and to chat using voice recognition. “TJBot is an open source project IBM is putting out as an easy way to access Watson AI services in the cloud, including speech to text, and text to speech,” said Yasaman Khazaen, research staff member with IBM Research in Yorktown Heights. For example, “You can ask it, what’s your name? and it will respond.” TJBot was announced in November.

This content is original to AITrends.