Deep Green Helps Warriors Plan Ahead

November 2007
By Henry S. Kenyon

The Defense Advanced Research Projects Agency’s (DARPA’s) Deep Green program is developing an automated decision aid for U.S. Army and Marine Corps commanders and their support staffs. It will help commanders keep ahead of enemy forces’ decision loops by prompting officers to make decisions at key times.
Decision-making aid tracks possible mission outcomes, prompts officers to develop contingencies.

Uncertainty has challenged military operations since the days of the ancient Greeks. An experimental decision-making technology could help future commanders see through some of the fog of war by helping them plan operations, recognize when a plan is not working and develop alternatives to keep ahead of the enemy.

The Defense Advanced Research Projects Agency’s (DARPA’s) Deep Green program seeks to develop an automated system to help officers and their staffs quickly make decisions and generate options. Another key goal of the effort is to replace the observe-orient-decide-act loop with a new paradigm. Deep Green Program Manager Col. John Surdu (USA), Arlington, Virginia, notes that when he served in an infantry brigade headquarters, rapidly changing events would often overturn operational plans before the plans could be implemented. 

Instead of having a planning phase followed by an execution phase, Deep Green will execute both phases simultaneously. The idea is to allow commanders to think creatively about the options available to accomplish a mission. The software-based system then takes these alternatives and analyzes how they might play out to help commanders stay within an enemy’s decision cycle. The technology will allow commanders to generate options rapidly and proactively to avoid any surprises, which the colonel describes as a point in a battle where a commander has no options. “The assertion behind Deep Green is we should be surprised much less frequently,” he says.

Commanders and their staffs generate mission options. The software examines the options and creates a variety of possible outcomes. Col. Surdu notes that simply having an A and B alternative does not create two possible futures but many possible outcomes. “If we take many, many options for both sides, it [the software] starts to generate lots and lots of possible futures—more than humans can keep in mind,” he observes.

This overabundance of options is why only one or two courses of action, perhaps with a few branches or alternatives, are traditionally prepared. The colonel explains that the software will generate outcomes by analyzing data from the ongoing operation to determine which of the previously created futures is becoming more likely. “Instead of just being able to generate a couple of options, we can create a host of options with lots and lots of branches and sequels. No matter what direction the enemy takes us during an operation, we still have options. We’re not caught flat-footed,” he maintains.

Deep Green originated from conversations with warfighters. A common theme among them was that many plans were flawed or failed before an operation was ever launched. Commanders had to go into a reactive mode immediately because the enemy did something that was unexpected. Deep Green seeks to bring any possible enemy actions to light and to have pre-generated responses ready for commanders.

Col. Surdu envisions that all echelons will use Deep Green. However, the focus for the DARPA program is to develop a system to support commanders at the brigade and battalion levels. Deep Green also can be applied at echelons above brigade, but the colonel notes that some important considerations must be addressed before the system is used at lower echelons. “A company commander’s fight tends to be more reactive than the kind of deliberate planning that occurs at the brigade level,” he observes.

Deep Green’s software is intended to have a modular architecture. For example, it may be possible in the future to remove a brigade-focused module and replace it with a company-focused module. The DARPA effort also is examining full-spectrum operations. The program’s first phase will support mid-intensity conflicts and combat or kinetic aspects of low-intensity conflict. Phases two and three will support three-block war options and noncombat humanitarian and peacekeeping missions.

Mission options can be asynchronously generated whenever a commander or staff officer requires them. Users can run drills at any time. Deep Green will collect data constantly from the current state of an ongoing operation to update metrics associated with the generated futures. As events occur, some futures become less likely. Col. Surdu notes that if a particularly good future is becoming eclipsed by a bad outcome, the system will detect this shift and alert commanders to generate options that may move the mission back to the desired outcome.

At critical times during an operation, the software will focus commander and staff attention to make decisions or build options. The colonel notes that there is a difference between futures and options. A mission may be heading toward an undesirable future, but the options to change that state are different from the predicted future. However, he relates that until the software has been developed, he does not yet know how far in advance the system will be able to detect a bad future and provide warning. “That’s one of the program’s performance metrics—does it give enough warning so that the commander and the staff are more effective with the system than without it?” he shares.

Col. Surdu notes that much ongoing research is currently focused on computer systems that generate options for humans to choose from. With Deep Green, humans will generate options and make the decisions, but the software will help them explore how those options may play out and help focus attention on areas where alternatives need to be created and decisions must be made. Another goal is to present decisions in such a way that commanders and staff understand the ramifications of their choices. “A commander knows that if he blows up a bridge, the first order effect is that nobody can cross. It’s the second and third order effects, such as we can’t provide logistics to our forces or people can’t use that bridge to get to work, that the computer will present to the commander so that he can make a more informed choice,” he says.

The software-based Deep Green system will reside in battle command systems such as the Command Post of the Future and Blue Force Tracker where it will model the possible outcomes of commanders’ operational choices and contingency plans.
Deep Green is divided into several components or applications: Commander’s Associate, Blitzkrieg and Crystal Ball. The Commander’s Associate tool is the primary interface between human users and Deep Green. Its interface allows users to draw freehand on it and to speak instructions to generate options. In the future, it will be able to infer or deduce commanders’ intents from their sketches and speech. The goal is for officers to draw a course of action on a monitor the way they would on a piece of paper and to have the computer understand the drawings and generate futures. The Commander’s Associate also will be able to prompt commanders at times when they need to generate options. “Visualization happens in a commander’s head, not on a screen. We have to figure out how to present that information in a way that helps the commander visualize the spatiotemporal aspects of the operation,” Col. Surdu says.

The next major application is Blitzkrieg, which takes the options generated for friendly, allied and neutral forces and models possible futures. “The intent behind Blitzkrieg is to be a completely different paradigm in military modeling,” Col. Surdu maintains.

To generate a range of options with current technologies, a simulation must be run hundreds or thousands of times, and then the outputs are placed in separate categories representing different possible outcomes. Col. Surdu shares that most combat models are built to favor the middle of the statistical curve. With Deep Green, Blitzkrieg will generate a broad spectrum of qualitatively different futures. For example, if two forces encounter each other, current combat models would run a series of attrition calculations and would present commanders with a statistical distribution associated with casualty rates and outcomes. With Blitzkrieg, if two forces encounter each other, one outcome may be that one side is rapidly defeated or that one side disengages as it begins to lose, and the winning side must choose whether to pursue. Another option may be that both sides suffer heavy casualties, withdraw and regroup. 

Blitzkrieg uses heuristics and quantitative techniques to assess the relative likelihoods of the outcomes. Then for each individual outcome, it will continue to simulate options until it encounters terminating conditions.

The third major component of Deep Green, Crystal Ball, serves several functions. It controls how Blitzkrieg generates rules, and it monitors data from an ongoing operation and updates the likelihood metrics associated with possible futures. This capability allows the system to identify decision points where commanders need to make choices, and it also identifies those futures where commanders and staffs need to generate additional options. A program goal is to design Crystal Ball so that the software can learn to predict the potential outcomes of futures better, based on data from the ongoing operation.

Deep Green also has an integration function that arranges all of the systems and specifications for passing data within the system and that ensures all of the modular components can plug in. The integration function also will help component developers integrate their sections with existing U.S. Army battle command systems. Col. Surdu says that the ability to have modular components in the system not only will allow the system to serve different echelons, the architecture also will permit new technologies or systems to replace Deep Green without difficulty. “If 20 years from now, people are using Deep Green but none of the code that we wrote is in it anymore, that’s a success,” he says.

But Deep Green must overcome several technological challenges to be successful. One area concerns the speech and text recognition systems required for Commander’s Associate application. Col. Surdu notes that a recent pilot effort demonstrated that the sketch recognition capability is feasible. Another aspect of the Commander’s Associate application that has not been attempted before is making the software understand a commander’s plan from this collection of symbols. The colonel notes that there are some promising artificial intelligence programs that may make this capability possible, but he cautions that this technology remains a major hurdle. 

Col. Surdu expects the program to be underway by the second quarter of the 2008 fiscal year, after all proposals have been accepted. The program’s three phases each will span approximately 12 months.

To enter within an enemy’s decision loop, Deep Green’s objective in the first phase is to generate three options for allies and enemies, conduct all analysis and present the data back to a commander in less than 30 minutes. By the end of the program’s third phase, the goal is to reduce this time to three minutes.

DARPA is working with the Army’s battle command development community to coordinate this effort. Col. Surdu notes that he is working with the Program Executive Office, Command, Control, Communications, Tactical (PEO C3T), Fort Monmouth, New Jersey; the Communications–Electronics, Research, Development and Engineering Center (CERDEC), Fort Monmouth; and the Battle Command Laboratory at Fort Leavenworth, Kansas. “Once we demonstrate the efficacy of this approach, the intent is to turn the technology over to either CERDEC or Project Management Battle Command to integrate this into battle command systems of record,” he shares.

Deep Green will integrate into existing battle command systems. The colonel notes that the target system is the Command Post of the Future, or CPOF (SIGNAL Magazine, June 2002). He explains that the CPOF has a built-in collaboration infrastructure that will permit users in different physical locations to operate the Commander’s Associate tool simultaneously. The CPOF’s interactive technology allows tools such as collaborative sketching and information sharing to operate. Deep Green will ride on this system’s infrastructure. “My guess is that Deep Green will be an icon, a tool, on a CPOF display,” he says.

Web Resource
DARPA Deep Green Federal Business Opportunities:


Enjoyed this article? SUBSCRIBE NOW to keep the content flowing.