AARON B. FRANK, RYAN ANDREW BROWN
How Advanced Research
Project Agencies Pick
Their Programs
The Benefits and Limits of Goldilocks
Problems
T
he U.S. government has increasingly looked to employ Advanced Research Project Agen-
cies (ARPAs) to advance scientific knowledge and develop novel technological capabilities
in a variety of departmental missions.
1
These ARPAs differ from traditional research and
development (R&D) organizations in that, by design, their mandate is to fund high-risk,
high-reward programs that, if successful, would provide the United States with transformative capa-
bilities.
2
These organizations—the Defense Advanced Research Projects Agency (DARPA), the Intel-
ligence Advanced Research Projects Activity (IARPA), the Homeland Security Advanced Research
Projects Agency (HSARPA), the Advanced Research Projects Agency-Energy (ARPA-E), the
Advanced Research Projects Agency-Health (ARPA-H), and Advanced Research Projects Agency-
Infrastructure (ARPA-I)—take on scientific and technical challenges within an organizational
framework that empowers a small number of program managers (PMs) to make substantial invest-
ments in research areas to test whether paradigm-changing approaches to meeting their depart-
ment’s mission needs are viable and to identify gaps in scientific and technical knowledge that limit
what is possible.
To better understand what types of risks ARPA programs should take, we held discussions in
March through June 2023 with 18 experts from within and outside of RAND, using the concept of
“Goldilocks problems” as an initial framing for our discussions.
3
Within this pool of experts, six
were current or former ARPA PMs or senior managers, and an additional four worked closely with
the ARPAs through their professional roles in other government research organizations or feder-
ally funded research and development centers (FFRDCs). The remaining eight experts have exten-
sive experience in the development, evaluation, and transitioning of technologies into national
security applications.
The premise of Goldilocks problems is that they are neither too easy, as to not warrant the use
of finite ARPA resources and attention, nor too hard, expending resources against unattainable
Research Report
