SRI International, which spearheaded the CALO (Cognitive Agent That Learns and Organizes) intelligent assistant for DARPA (Defense Advanced Research Projects Agency), has had more than one semantic project up its sleeve. One of them was Tempo.AI, which was spun off by SRI at the end of 2011. Earlier this year, the smart calendar app for the iPhone was formally launched, with Thierry Donneau-Golencer as co-founder and AI lead.
Donneau-Golencer, having also worked on CALO, clearly has a strong history of work related to dealing with information and how to make sense of it. “A lot of it had to do with semantic analysis, deriving meaning and useful information from content,” says Donneau-Golencer, with Tempo representing the next step in smart search across content by making the job more proactive.
At October’s Semantic Technology & Business Conference in NYC, Donneau-Golencer will share with attendees insights into the role semantic technology has in helping find and correlate information for users, with the least input possible required.
Tempo, for example, was built to be an assistant that gets smarter with more data and use, learning patterns and behaviors and constantly optimizing the smart calendar experience so routine tasks take the fewest taps. The artificial intelligence behind it pulls together a variety of features for users, including opening relevant documents and emails for meetings that are on the calendar for easy access; providing a one-step way to alert other meeting attendees that they’re running late; or offering single-step dialing into conference calls because Tempo can recognize such objects and handle the pass codes as well as dial-in numbers. Users also can use Siri to verbally clue Tempo in to meetings to put on their calendars.
The main point of the talk, he says, revolves around acknowledging that users have a lot of personal content that tends to be scattered everywhere, often in unstructured and not easily searchable, findable, or machine-understandable form – and then he’ll discuss how to do something about that. “How do you extract all this information, use it and make it useful?” says Donneau-Golencer. “That’s the actionable part, and Tempo is an example of a growing number of applications to do this.”
Semantic tech is critical to making such apps work, he says, “because a lot of the information is buried inside sources in unstructured ways. Plus [there are] other applications, like being able to search and find information — everything is around semantic search. Also you need to personalize such apps, you need a user profile specific to you, which means you need to make sense of the data.”
Because such systems are personal and have to be actionable, they have to constantly adapt to new information, to the user context. “The Knowledge Graph,” he explains, “has to constantly keep apace. So the system has to learn from you and your environment.”
The trickiest part of that, he says, is the concept of transfer learning. That is, is there anything that can be learned from someone else or from some specific application or context that can be transferred to another because it has applicability to that person and their environment, too? “Tempo, for example, may have learned a lot about me that also can help my colleague, but how can we transfer that information – what makes sense, where does it break down, and what about privacy?” he asks.
That said, Donneau-Golencer also thinks that privacy is becoming less of a concern among users now accustomed to greater disclosures of information, with the advent of online social networks. “People share more than we would have imagined,” he says. At Tempo’s launch, for example, people were connecting a lot more data to the cloud-based service than had been expected. Says Donneau-Golencer, “It seems as if people are willing to share more when they see value.”
To hear more from Donneau-Golencer at the upcoming conference, register here.