Microsoft Chases General AI with New Research Group
Microsoft today unveiled Microsoft Research AI, a new research and incubation hub that’s slated to take on some of artificial intelligence’s biggest challenges, including the creation of a general AI, the field’s Holy Grail.
Microsoft Research AI, which was officially unveiled at a London event today, will look to push the envelope in a range of fields connected to machine learning, deep learning, and AI, including computer vision, natural language processing, human-computer interaction, and robotics. It will also look to unite these various sub-fields in pursuit of a more useful and general AI that benefits Microsoft customers and society at large.
Microsoft is no newbie to AI. Just 10 months ago, the software giant formed Microsoft AI and Research, a group that brings together more than 7,000 computer scientists, engineers, and researchers from Microsoft’s various labs. However, the company does not have as high a profile as some of its competitors, including Google and its various initiatives, including Google Brain and Google Research.
“A key focus for this initiative,” the company says, “is to probe the foundational principles of intelligence, including efforts to unravel the mysteries of human intellect, and use this knowledge to develop a more general, flexible artificial intelligence.”
The phrase “general AI” refers to the longstanding goal of AI researchers to create computer programs or robots that mimic human intellect in a nearly seamless manner. Despite all the recent advances in the field deep learning, the various sub-fields are largely evolving separately.
“Another core goal of Microsoft Research AI,” Microsoft AI and Research Group executive vice president Harry Shum writes in a blog post today, “is to reunite AI research endeavors such as machine learning, perception and natural language processing that have evolved over time into separate fields of research. This integrated approach will allow us to develop sophisticated understandings and tools that can help people do complex, multifaceted tasks.”
Google is taking a similar path with its research. Just last month, a group of Google researchers published a paper describing its MultiModel architecture that unites various subfields of AI and machine learning.
Microsoft Research AI will house a core of about 100 data scientists and researchers who are currently working across on 15 groups, including deep learning, conversational systems, machine teaching, machine learning and optimization, and NLP, among others. The group’s websites lists 13 current projects, including aerial informatics, biomedical NLP, immersive rooms, and others.
Also on the group’s docket is the emerging field of machine reading, which Shum says “has incredible potential,” such as helping a doctor quickly find pertinent information among thousands of documents. But again, doing this well requires the capability to cross-train algorithms and models that currently are siloed.
“To do machine reading well requires combining AI disciplines such as natural language processing and deep learning,” Shum writes. “Our leading work in machine reading also is an example of the progress we are making toward a broader goal we have at Microsoft: creating technology with more sophisticated and nuanced capabilities.”