Innovation

Mastery of AI has been 'harder than expected' and 'future is uncertain,' says Microsoft's AI chief

Microsoft's head of research, Eric Horvitz, highlights advances in AI and concern over its future applications.

Eric Horvitz
Eric Horvitz speaks at MIT's EmTech conference
Image: Hope Reese/TechRepublic

"I like to think about advances in AI through lens of the history of flight," Eric Horvitz, director of the Microsoft Research lab told a crowded room at EmTech, MIT's annual technology conference in Cambridge, MA. "It started as a few guys on a windy beach looking at a kite," said Horvitz. "And 50 years later, we have a 707."

Kicking off the conference, Jason Pontin, editor of the MIT Technology Review, had asked the audience: "what do we want from smart machines?" It is up to us, he stressed, to determine the parameters for our new technology—how it should be applied, and how it can help us.

Horvitz opened the discussion by addressing this question, highlighting the history of augmented knowledge and how it is becoming woven into our natural environments. As the former president of the Association for the Advancement of Artificial Intelligence (AAAI), the world's largest scientific society of AI researchers, Horvitz has long been interested in the applications of machine intelligence in areas like healthcare, transportation, and communication. As director of research at Microsoft, Horvitz is helping lead the revolution in AI research.

"It's important to take stock of recent advances," he said. Today, there are 1200 people at Microsoft research. A quarter, he said, are "solely focused on AI." New leadership at Microsoft, Horvitz noted, is partly to thank for this. "There is more focus on research than in the past," he said. "There are new forms of joint ventures."

"There's a revolution in probability," Horvitz said. "A revolution in machine-learning." This revolution, according to Horvitz, has been "fueled by almost limitless storage for capturing data."

What is new in AI? Horvitz said there's been a "renaissance in machine-learning as core pillar in AI." A lot of it, he said, comes "under the hood," in predictive models, and in pathways from A to B. Other great areas of advancement are automatic machine reading techniques and "personal empowerment tools" that will help people "plan, remember, know where they're going."

Horvitz was clear that while there's been "fast-paced progress" in AI, "it's unclear how fast things will move with machine intelligence," he said. "The mastery of AI has been much harder than expected." The important consideration, he believes, is to move towards "the more deep general intelligences," rather than simply building "narrow, deep wedges of intelligence" that have been a primary focus up to this point.

While we have "a lot of competency in pattern recognition," Horvitz said, we need to focus on "weaving together the new competencies we're seeing." We need to "bring together a symphony" with our new tools, "building a system where the whole is greater than the parts.

"I'm pretty sure that the next leaps in AI will come from integrative systems, rather than wedges," he said.

Horvitz did not hide his concerns about "the rough edges" of AI. "The future is uncertain," he told the crowd. "Will systems push aside human beings? Eliminate jobs? What are the socioeconomic implications?" History "can tell us how complex things are when society faces new types of automation." We saw a "really complex panoply of possibilities when machines began overtaking humans on assembly lines," he said. Still, he believes that it's "just as likely we'll have new jobs and new collaborations with machines as displacements at work," Horvitz said.

Still, Horvitz said that "it's important for computer scientists and others to take concerns seriously and address them. Let's push them hard, let's understand them better. How can we lessen worries?

"We need to address [them] maturely and explore deeply whether there's something there or not," Horvitz said. And we need to consider: "How can technologies be applied in ways we don't expect?"

Also see

About Hope Reese

Hope Reese is a Staff Writer for TechRepublic. She covers the intersection of technology and society, examining the people and ideas that transform how we live today.

Editor's Picks

Free Newsletters, In your Inbox