Disruptive by Design: Policy Lessons for Building a Multidomain AI Ecosystem
When Google announced it was acquiring Nest for a little over $3 billion in 2014, analysts thought the company wanted to enter the home appliances market.
It was all about the data.
Google gained access to a treasure trove of information about consumer demands for heating and cooling. The company learned when people turned on their furnaces and shut off their air conditioners. Google could pair this information with the type of household, neighborhood and city.
By collating this information, applying predictive analytics and then later, machine learning models, Google offered new insights into utility companies. It could predict the outcome of weather patterns, utility price fluctuations and household income on likely utility usage rates, and it found new customers eager for this information but unable to gather it themselves.
Innovative information use cases also exist across the U.S. Department of Defense. Registering and prioritizing these information requirements can help determine what data should be actively harvested. The department might find the following approach to cost-benefit analysis useful.
First, approach decisions on an artificial intelligence (AI) ecosystem with a wide lens. Look across the entire enterprise and determine how information gathered within one domain, one service or one operational function may be harnessed throughout the force. Sensors on sailors may reveal data helpful for soldiers and vice versa.
Second, define the value of the information to decision makers. Knowing what information matters most helps analysts provide objective frameworks to examine the pros and cons of various data-gathering platforms and analytical methods.
Next, determine whether the information makes the organization more effective or efficient. Would collecting and analyzing certain data sets increase the likelihood of mission success or reduce costs?
Also, assess implementation costs holistically. Measure both the reduction in costs associated with previous forms of information gathering with the costs of adding new tools and new support personnel. And consider how a new information system may introduce new vulnerabilities.
Lastly, conduct a sensitivity analysis to determine what information matters most to the overall value of a decision. Factors may vary due to an underlying relationship with another variable. Knowing what is at the heart of an observed phenomenon requires sophisticated analysis.
The Defense Department needs to look hard at where it can gain the greatest benefit at the least cost in implementing an AI ecosystem. By applying lessons learned from previous endeavors—and by approaching the engineering decisions of such an ecosystem holistically—the Defense Department can maintain its competitive military edge. The time has come to move from speculating about use cases for AI to actively building an implementation plan.
Maj. Ryan Kenny is an officer in the U.S. Army Signal Corps. He is a chief of staff of the Army-sponsored Advanced Strategic Planning and Policy Program Goodpaster Scholar. He is currently pursuing a doctorate within the Department of Engineering and Public Policy, Carnegie Mellon University, Pittsburgh, where he is focused on policy concerns of artificial intelligence and decision making. The views expressed here are his alone and do not represent the views and opinions of the U.S. Defense Department, U.S. Army or other organizations with which he has had an affiliation.