Machine Learning Could Help Chip Away at the Security Clearance Backlog
Augmenting human investigators is a better option than doing reinvestigations less frequently.
At the end of July, the Pentagon announced a change to the time period for conducting background investigations to help reduce the huge backlog of people waiting for their government clearance. It plans to simply do reinvestigations less often, officially stretching the process from five to six years, when in reality, it can already take much longer than that. While this may free up some resources to conduct initial investigations, it is a partial solution at best.
The government is exploring other options to fix the clearance process. For example, the White House wants to reorganize the agencies that perform clearance investigations. The Senate Intelligence Committee approved the Intelligence Authorization Act for fiscal years 2018 and 2019, which includes measures that would promote information sharing between federal agencies and would establish a governmentwide policy for granting interim clearances. Pilot programs are attempting to move the process away from today’s slow and labor-intensive process in favor of continuous evaluation that incorporates data-mining software that can catch clearance issues in real time.
These are steps in the right direction, but the problem is a crisis right now. The backlog jumped to 740,000 people and the investigations are taking even longer.
The good news is that the commercial sector already uses technological solutions that could dramatically reduce the time needed to conduct an investigation and at the same time, make the screening process far more effective than it is today.
The information collected during background investigations are essentially large data sets collected from the person seeking the clearance. This data, much of it collected on yellow legal pads during the investigator interviews, is then manually compiled and reviewed. Any abnormalities such as an arrest, unexplained wealth, or unusual contact with foreign nationals must be identified and flagged by the investigators, prompting more review and investigation.
Even basic machine-learning technology could save countless hours. Using historical data on the millions of clearance reviews that have been conducted and the counterintelligence case profiles of people who misused classified information, machine-learning models could be trained to identify the types of discrepancies that cause clearance rejections and quickly screen new requests. This comprehensive review of data sets is something that no investigator could hope to achieve.
Anyone who has clicked on something in the “Recommended for you” section of Amazon or Netflix has used machine-learning technology. The recommendations are based on computers reviewing datasets of selections from millions of people and comparing them to your choices. Can you imagine how long and expensive it would be for Netflix to send an investigator to your house to interview you about your movie preferences, then talk to all of your friends to confirm your answers, record them by hand on a legal pad, then type them up, review them with a panel at Netflix headquarters, then generate a recommendation and send it to you?
Obviously, the stakes are much higher when we are dealing with security clearances. False positives could lead to the rejection of qualified candidates for sensitive jobs, and false negatives could miss the red flags that should prevent a candidate from getting a clearance. However, these are arguments for augmenting the work of investigators with machine-learning technology instead of relying on humans alone.
It is important for key decision-makers at the Pentagon to understand that machine learning is a tool that investigators and reviewers should use in making clearance determinations by flagging risk factors to be further evaluated before a final determination is made. Using machine-learning tools does not mean simply trusting AI robots to make critical national security rulings, but instead help more efficiently flag risks for human evaluation.
The outdated clearance process is horribly inefficient, costing taxpayers millions of dollars and preventing qualified candidates from doing critical national security work that our nation relies on. Most importantly, the manual process of conducting investigations with no follow-up for years is far less effective at identifying potential issues than an automated process that could identify these potential issues in a matter of minutes and flag them for further investigation.
The security clearance backlog is a crisis, and the time for reform is right now. We can’t afford to wait years for reorganizations and pilot programs to begin to bear fruit.
Erin Hawley is the vice president and general manager of public sector at DataRobot.