After conducting speech research in laboratories at Purdue University, the University of Maryland, and Johns Hopkins University, Dr. Mary Harper’s quest to build advanced speech recognition systems led her to the Intelligence Advanced Research Projects Activity (IARPA).
“As you begin to engage larger and larger groups of researchers, which is something that we get to do here at IARPA, you can actually solve much more interesting and important problems than is possible as an individual researcher,” said Harper, who leads Babel, an ambitious IARPA program that has collected speech for 26 of the world’s 7,105 living languages.
Some of the languages are as well known as Cantonese and Pashto, while others, such as Zulu and Igbo, are more obscure. Harper’s mission: Use data in those languages—collected in a variety of recording conditions including cellphones in a crowded room or microphones in a reverberant room—as the foundation for technology that would enable computers to be trained in a week or less to transcribe speech in these languages to text.
Harper belongs to a cadre of program managers (PMs) that are the backbone of IARPA. The PMs are hired on three- to five-year contracts from academia, industry, national laboratories, and federal agencies to pursue high-risk solutions with the potential for high reward. Upon completion of their contract, some PMs return to their former positions, while others move on with the addition of IARPA to their resume.
At IARPA, new PMs take on programs orphaned by departing PMs while also embarking on new programs of their own—knowing the programs they launch will likely be completed by someone else. It’s a cycle that keeps fresh ideas and approaches flowing through the building, located in a research park on the University of Maryland campus.
IARPA was conceived in 2006 and officially stood up in 2007 under the Office of the Director of National Intelligence to support the Intelligence Community (IC). Built on the model of its older and more commonly recognized cousin, the Defense Advanced Research Projects Agency (DARPA), IARPA has unclassified programs addressing computing, global-scale 3D modeling, biometrics, forecasting, cybersecurity, and geospatial and signals intelligence—to name a few. The end goal is to successfully transition research to federal agencies for intelligence gathering and processing.
Though DARPA and IARPA serve the Defense Department and the IC, respectively, the two organizations share information to avoid redundancy. For example, Babel leverages research from and expands upon earlier DARPA research. IARPA programs seek to solve problems so complex they require substantial resources only an organization such as IARPA can offer.
“People say that in order to be successful in an IARPA program, you’re looking at one or two miracles to happen,” said Dr. Hakjae Kim, who is on assignment to IARPA from the National Geospatial-Intelligence Agency (NGA). Since arriving at IARPA, Kim pitched CORE3D, a global-scale 3D modeling program that would use data from several satellite sensors to add a third dimension to map features through automation—taking humans out of the loop.
IARPA programs seek to meet a long-term need of an intelligence agency or the overall IC. The requesting agency offers consulting before a program begins, representatives for the program’s advisement team, and a plan to transition a viable solution into the agency for field use.
“At every opportunity, we try to make sure there’s somebody who’s going to benefit from this research,” said IARPA Deputy Director Dr. Stacey Dixon, who was most recently at NGA. “[The agencies] have an opportunity to influence the types of metrics we set, so we know whatever capability we’re trying to develop translates into something they actually need accomplished.”
Dixon has held several positions throughout her career, including serving as a professional staff member on the U.S. House of Representatives Permanent Select Committee on Intelligence. In this role, she helped with the legislative groundwork that established IARPA.
Leading IARPA since August 2015 is Director Dr. Jason Matheny, who joined IARPA from Oxford University, where he was researching probability theory and risk assessment.
“I came [to IARPA] because I could fund 100 people who were world-class scientists and engineers who could work on those problems,” said Matheny, whose first IARPA position was to manage forecasting programs, one of which supported a research effort that alerted U.S. public health officials to the West African Ebola outbreak.
Matheny served in various posts on his way to the director’s chair.
“I got to multiply my own research effort by a factor of 100,” Matheny said. “I know some program managers who came from academia or industry labs or other government labs are here because of the ability to create an entire research community that didn’t exist before.”
No Pet Programs
Program managers shepherd programs from conception, through a “Proposers’ Day”—a conference in which the research community weighs in on the idea, and the writing and releasing of a broad agency announcement to inform potential researchers around the globe of the IARPA opportunity.
The PM job is heady. For example, the research groups PMs manage might include Nobel Laureates or McArthur Genius Grant recipients. Cold Spring Harbor Laboratory, part of a consortium that signed a $21 million IARPA contract in February to map the brain as part of the Machine Intelligence from Cortical Networks (MICrONS) program, led by Dr. R. Jacob Vogelstein, has eight such prize winners. MICrONS is part of a neuroscience effort that seeks to make computers think more like humans by reverse-engineering the brain’s algorithms. Dr. David Wineland shared the 2012 Nobel Prize in physics partly for his IARPA research on quantum computing.
The fresh talent circulating through IARPA is, in part, the result of a saying by Matheny’s predecessor, Dr. Peter Highnam, who is said to have frequently reminded PMs not to fund their “pet rocks.” Highnam now directs research and development at NGA, but the term “pet rocks” lingers as a reminder to PMs that their programs can end at any time if adequate progress is not made. This is why approximately a quarter of the budget for an IARPA program is allocated to testing and evaluation—for a typical unclassified program, this is around $2.5 million per year on “T&E,” according to Matheny. The overall IARPA budget is classified, as are some of its programs.
Approximately 20 percent of IARPA programs and as many as 50 percent of its research efforts end short of a solution, Matheny estimates. Programs and research efforts are cancelled for a variety of reasons that do not always equate to failure. And cancellations aren’t kept under wraps; in fact they demonstrate that IARPA is hard at work, Matheny added.
“If we had a failure rate of only 10 percent, I would say we’re picking problems that are too easy,” he said.
In addition to taking on existing programs, IARPA PMs are hired based upon their own pitch of a new program. PMs pitch their program ideas by answering “Heilmeier Questions,” created by Inventors Hall of Fame engineer George Heilmeier. The questions ask for a clear, jargon-free explanation of the problem and its current state of solution, as well as an explanation of what IARPA can add to that state and what it would cost to do so.
Seemingly straightforward, the Heilmeier Questions often stump aspiring PMs.
“I came in and pitched an idea to the director (then Dr. Lisa Porter),” said IARPA PM Dr. David Moehring. “I think her quote was, ‘It’s clear you are technically competent, but you need more practice in organizing and managing milestone-driven programs like those at IARPA.’”
Moehring refined his Heilmeier answers, improved his pitch, and was later hired. His newest program, LogiQ, seeks a solution to the problem of encoding imperfect physical qubits into a logical qubit. Qubits—or quantum bits—are somewhat like the analog computer’s bits. But qubits, the foundation of quantum computing, can greatly shorten the time needed to decipher encrypted messages, which is one reason the IC wants to leverage quantum computing. Moehring’s program aims to take the IC part of the way there in five years. The U.S. government has worked on quantum computing for 20 years and IARPA has several transition partners for LogiQ.
But somebody else will reap the recognition should LogiQ’s quest come to fruition. Moehring learned this lesson when he took over the Multi-Qubit Coherent Operations (MQCO) program from its original PM, Dr. Michael Mandelberg, who recruited Moehring as his replacement.
“At the end of the program, the people would informally call it ‘Dave’s program,’ which is nice because I did manage it longer than [Mandelberg],” Moehring said. “But it also gives me the realization that [LogiQ], which I started, will be somebody else’s program someday. So be it.”
This is why PMs often influence the selection of their successors. Dr. Chris Boehnen is in his first year at IARPA from Oak Ridge National Laboratory. He already knows he won’t see the end of his program, Odin, in which he seeks to teach a computer when it has been attacked by adversaries trying to thwart facial, fingerprint, and iris identification elements.
“My children have not yet reached the point of going off to college, but I expect it’s like that,” Boehnen said. “You’re happy to see them going, happy that they’re successful and doing well, but you’re probably a little sad that they’re not home with you any more.”
It’s the price PMs pay for working at IARPA, which offers opportunities to manage world-class research in support of the Intelligence Community, but tempers those offers with term limits that steer PMs away from pet rocks. That’s the nature of IARPA, as is the inevitable change that comes with science and technology.
“IARPA is sort of an evolving organism,” Matheny said. “Not only should our research change, our structure should change to adapt. The staff will be constantly changing, and I think that’s a good thing.” ￼