Selective Curiosity

IBM Fellow Jeff Jonas seeks ‘maybes that matter’

iStock-545793676

Standing on the grounds of the U.S. Capitol, one hears birds, buses, and the chattering of tourists. Off in the distance is a siren, the overhead whir of a helicopter, the clanging of a construction site. In the nation’s capital, these sounds are typical, and they generally don’t elicit inquisitiveness. Throw the crack of a gunshot into the mix, and one instantly becomes quite curious.

This is what IBM Fellow and USGIF board member Jeff Jonas has coined “Selective Curiosity.”

“You can’t be curious about everything,” Jonas said, noting that the out-of-the-ordinary cannot always invoke our curiosity either. “There are simply not enough free computational cycles to investigate everything.”

In his lab, Jonas is developing Selective Curiosity features, in which a machine will not only determine it needs more information to reach a conclusion, but independently seek the information from sources such as Google or LexisNexis.

In a 2014 blog post, Jonas wrote that his G2 Selective Curiosity tool—which he hopes to reveal later this year—“finds a maybe that would matter.” In the Intelligence Community, machines routinely produce “maybes.” In such situations, this new technology aims to help a machine determine what it wishes it knew (i.e., an individual’s employer or home address) and figure out where it should go to find the answer.

“In big data there are millions of maybes,” Jonas said. He added that tipping and cueing—which tips analysts or an algorithm toward a particular person or object in the data—is riddled with false positives. “Fixing these accuracy rates will not come from more algorithm work, it will come from more data.”

But which data? Jonas said Selective Curiosity will allow the machine to figure out—and then search for—exactly the type of data it needs, yielding significantly more accurate decision-making than existing methods can deliver.

Return to feature story: Machine Learning, Big Understanding

Posted in: Features   Tagged in: 2016 Issue 3, Data, Machine Learning & AI

Strength in Numbers

Open Source software is defined by the community as much as it is by the code

OSINT Goes Social

Social media presents new opportunities and challenges for deriving open source intelligence

Pigeonholed

Privacy policies threaten advances in geospatial technology