When the Defense Innovation Board first came into existence a year and a half ago, its chairman, Eric Schmidt, and his fellow panelists vowed that they would not be in the business of writing reports. There are enough federal advisory committees that do that sort of thing already, they reasoned.
That abstention hasn’t lasted long, because Congress does not share the Silicon Valley-centric group’s distaste for voluminous, paper-based descriptions of problems and how to solve them. The DIB’s first official tasking from Capitol Hill, directed in this year’s Defense authorization bill, is spend the next year writing a report on one of the Defense Department’s thorniest problems: software acquisition.
At their quarterly meeting last week, members vowed to take the project seriously, but suggested the end product is not likely to be a PDF document that analyzes and explores DoD’s track record on acquisition.
Again, there are enough of those, said Richard Murray, one of the board members who’s been tasked with leading the study.
“Can’t we just go out and essentially grab all of those data and do machine learning on very large data sets and pull together stuff that’s already out there, whether that’s PowerPoint files, or spreadsheets or printed documents? Machine learning can do that these days,” said Murray, a professor at the California Institute of Technology who specializes in networked and autonomous systems. “Once we ingest that data, we could use modern data analytics techniques to try and get insight about things that are already being done. What are the types of features of a software acquisition program that make it really go wrong, versus those that seem to be working well?”
In theory at least, the board should have no trouble getting its hands on the large data sets it hopes to use for the study. The same section of the NDAA that ordered the board to deliver its report also directed the secretary of Defense to give the DIB “timely access to appropriate information, data, resources, and analysis.”
But Murray emphasized the board wanted to do all it can to eliminate the need for manual data calls — the type of personnel-intensive information requests that tend to inform traditional reports. Rather, it’s more interested in raw data. And the product the DIB winds up producing might end up being helpful to the acquisition personnel who deliver the data, rather than just being a long read.
“Maybe what we end up coming out with, instead of a report, is a deep neural network that will answer questions, like, ‘Here’s the set of requirements for a piece of software. How late is it going to be, and how much over-cost is it going to be?’ I don’t know, but we do want to think about this differently,” he said. “In order to get these techniques to work, we’re going to need lots of data — raw data — in forms we can process that way, and we’re looking for ideas about how we can do that.”
The board’s search for new study techniques stems, in part, from the incredibly expansive mandate Congress gave it.
The NDAA challenged the panel to review all of the DoD regulations that have a bearing on software programs, examine a cross section of ongoing acquisition efforts based on their “application types, functional communities, and scale,” identify best and worst practices throughout the department, and then deliver a set of recommendations to make software procurement and development more streamlined, rapidly adopt new technologies and improve the expertise of the DoD workforce involved in software acquisition.
The study will need to take into account programs that DoD and its contractors build from scratch in order to support mission-critical weapons systems, business software the department buys straight off the shelf, and everything in between, said Dr. Michael McQuade, another board member who is leading the effort.
“It’s our working assumption that how one might offer suggestions on acquiring an ERP system for personnel records is a very different object that people who want to find software that can change an air tasking order at a forward operating base,” said McQuade, the senior vice president for science and technology at United Technologies Corporation. “There’s not going to be one solution that fits all.”