Yahoo Launches $10 Million Research Effort to Invent a Smarter Siri
Academics at CMU will try to extend Yahoo’s services with mobile apps that act like personal assistants.
Mobile apps that can carry out even basic conversations could transform how we use our devices.
Yahoo’s future mobile apps could come with a dash of intelligence that lets them act like personal assistants. The company has made a $10 million grant to Carnegie Mellon University for a project called InMind, which is intended to create and test assistant-style services for mobile devices.
Under the five-year project, announced last week, researchers at the university’s machine-learning department will get access to the technology running Yahoo’s current services, such as mail and news, and will be able to develop new applications to extend them. Those experimental apps and services will be tested by Carnegie students and staff who allow them access to their Yahoo accounts, a group potentially numbering thousands of people.
“InMind will be the next generation of personal agent that will guide us in our personal lives, on our personal devices,” said Subra Suresh, Carnegie Mellon’s president, at an event on campus last Wednesday to announce the partnership.
With this project, Yahoo is asking for a more defined outcome than is typical for Internet companies funding academic research. The results will be similar to the kind of beta product more typically developed inside a company. This reflects the convictions of Yahoo CEO Marissa Mayer, who has said that using data to better serve the billion people who use the company’s products each month is crucial to rebuilding its business and reputation (see “Yahoo’s Expanded Labs”).
Ron Brachman, head of Yahoo Labs, says that he expects the InMind project to experiment with apps that are capable of rudimentary conversation—for example, asking a person follow-up questions and making suggestions based on new information. “This is missing from Siri,” he says, adding that although Apple’s personal assistant is impressive, it doesn’t attempt to understand the context in which it is being asked a question: it doesn’t understand what the user is doing or might need at the moment.
Brachman jointly conceived the focus of the InMind project with Tom Mitchell, head of Carnegie’s machine-learning department. Both men were involved in a Pentagon-funded project called CALO, which created the technology behind the startup that developed Siri and was later acquired by Apple (see “Intelligent Software Assistant”).
Justine Cassell, a Carnegie Mellon professor who will co-direct the InMind project with Mitchell, said at the launch event that the data Yahoo has on its users, from the contents of their e-mails to the records of their clicks on news articles, could inform new apps with a deeper understanding of their users and the world they live in. “In order to really personalize in an interesting way, you need to infer psychological states and beliefs,” she said.
Cassell pointed to existing Carnegie Mellon projects that might be used to build such apps. A system called NELL, for example, has spent the past four years trawling online text to build up a store of knowledge that could be used to identify a person’s interests. Cassell said the effort might also draw on CHORUS, an artificial assistant capable of conversation. Developed partly by a researcher now at Carnegie Mellon, it is based on crowdsourcing (see “Artificial Intelligence Powered by Many Humans”).