IE 11 Not Supported

For optimal browsing, we recommend Chrome, Firefox or Safari browsers.

Innovation: What’s the Next Phase?

Why government managers need to know about machine learning.

In last December’s issue, I predicted that cloud computing, social engagement, government-as-a-platform and the Internet of Things would be drivers of institutional improvement and operational gains in 2015. Government Technology headlines from this year attest to significant advances made by many cities nationwide on these four fronts. The next phase of government innovation will lie at the intersection of these recent advances: To drive technological aptitude forward, municipal governments need to dial down the lag between data collection, analytical output and well-informed action.

To this end, I expect that over the next 24 months there will be substantial advances in the importance of machine learning tools that will become clearer for some of the most forward-thinking city governments across the country.

Since machine learning is a computing technique that adapts itself to changing conditions, its most common application will be to make predictions. As machine learning programs are fed more data, they learn more, and so their predictive models become more precise and produce more accurate results. The concept is not new. Machine learning algorithms, for instance, underlie Google search and Siri voice recognition. But recent improvements in cities’ data aptitude — pioneered partly by a growing cadre of dedicated municipal chief data officers and chief technology officers — have unleashed a previously unimaginable capacity for advanced data analytics. As a result, machine learning is positioned to become a powerful management tool for municipal government.

Consider the current accountability crisis facing police departments across the country. Several high-profile incidents of police violence have eroded America’s trust in law enforcement, but imagine if supervisors could identify officers prone to overly aggressive conduct before a violent incident ever occurs and then use that information for recruiting, training and management purposes. An oversight system using machine learning techniques would make this possible.

The widespread implementation of body-worn cameras in police departments across the nation is producing immense archives of policing footage. While it would be impossible for employees to pore through these files, machine learning could analyze videos in aggregate, learning to detect subtle visual patterns that precipitate aggressive behavior. Combined with data from police reports, some of which inevitably belie implicit prejudices, the system could flag potentially aggressive officers and notify their supervisors. Such a program would automate the first step of better oversight by enabling more pre-emptive, surgical intervention by those in managerial roles.

Policing, of course, is not the only area of government that can benefit from machine learning. Gartner recently named advanced machine learning as one of the best strategic IT investments an organization can make. Researchers at MIT are currently investigating machine learning techniques that may reduce recidivism by giving parole officers a better statistical profile of repeat offenders. And IBM researchers are using machine learning to develop a system to predict pollution levels in Beijing 72 hours in advance.

Machine learning has its risks. A prediction is only as good as the data it’s modeled on — and if the information is rife with errors or biases, machine learning risks amplifying those errors into misguided future action. But if it’s implemented by intelligent data scientists who understand the social issues at stake, and used by smart managers who understand how the mechanics work, the value that these programs may unlock is immense.