Automated methodology makes the future more predictable.
A probability analysis program could enable surface and air military units to better predict a vehicle’s or a missile’s next move by discerning the likelihood that its track will either change or remain constant. Applying the same reasoning formula to study an entire mission, the system could combine factual and hypothetical data to predict the direction an enemy will take and produce theoretically sound solutions to tactically complex scenarios.
Current battlespace assessment techniques rely on information from various sensors to establish an accurate track on a moving target. As contact statistics are collected, tracking equipment forecasts a route based on a compilation of hard facts that reveal trends in an object’s actual speed and direction.
However, according to John D. Lowrance, program director for representation and reasoning, SRI International Artificial Intelligence Center, Menlo Park, California, standard military tracking devices do not account for unseen risk factors whose impact cannot readily be measured in an operational picture. These details can only be collected using intuitive methods that are outside the scope of sheer mechanization.
Based on evidential reasoning theory, SRI’s Gister technology examines not only confirmed information but also real-world data that is often uncertain, incomplete or inaccurate. This amalgam of fact and hypothesis produces a high-probability picture of future events. “The Gister concept focuses on what is not known about a situation using differing degrees of data reliability to construct a reasonable conclusion,” Lowrance explains. “To do this, several probability tests are performed on the available information to determine the weight each piece should have in constructing the big picture.”
Gister equipment collects sensor data that is interpreted using pre-established criteria to measure applicability to a mission. During this initial phase of battlefield intelligence preparation, the technology examines a number of environmental factors such as weather, vegetation and terrain. Using cartographic modeling, patterns in the landscape and their effect on weather conditions are considered to determine the positive and negative physical aspects of the battlespace. This information can help determine the best way to stay ahead of a threat.
The system analyzes the overall risk to personnel by applying an algorithm that synthesizes terrain visuals along selected routes. Aspects that either support or do not affect the operational picture are discounted. This sharpens the focus on elements that are detrimental to mission success.
Tactical threat detection occurs in parallel to environmental interpretation. This activity is the first step in the Gister evidential reasoning process. Sensor information in an initial report of a potential threat is held until the threat is confirmed to be real. As new data is added to the pool of evidence, groups of information are fused based on how well they substantiate a majority threat conclusion. Data that is less supportive of the mainstream indicators is de-emphasized. Information is grouped by similarity, which eliminates the neutralizing effect of dissimilar factors on overall battlespace clarity.
“One of the main challenges in dealing with multiple sensor outputs is determining the weight that each piece of data should be given in reference to the whole operational picture,” Lowrance indicates. “It is not unusual for sensory equipment to have varying locks on a target, which can raise the level of data ambiguity.” Gister’s goal is to lessen the impact of this ambiguity on the reasoning process by removing data from its original context. This limits the amount of influence that extraneous factors have on the main body of evidence, he adds.
Evidential reasoning incorporates data translation to convert multiple source mainstream information into a single language. As fused groups of data form, they are removed from the main body of evidence context. Highly probable factual material is separated from dependent hypotheses, allowing for a separate consolidation of confirmed and unconfirmed statistics. “This is the critical juncture when information important to the threat track or operational picture is clarified,” Lowrance explains. “In situations where strict command and control is required to refine a mission, differing pools of data certainty enable probability comparisons to be made in support of the optimal course of action.”
Time is a key element within the evidential reasoning theory. With the accumulation of target track data, patterns can be discerned to establish a projection template. Using compatibility relation methodology, mass data distributions are broken down into equal-sized segments that are referred to as frames. The Gister process then singles out repeated similarities in the information. A consolidation of frames at these points of similarity produces a more refined picture that emphasizes the probable direction that the object being tracked will move. “As patterns in data distribution are noted, probabilities about future performance can be predicted based on consistent variances in a track history,” Lowrance remarks.
In the next phase, Gister maps these data trends to formulate predictions of likely future patterns. The system fuses sensory information and incorporates the predictive analysis conclusions from any past trends to produce a picture of potential future behavior. This second fusion process melds data from the threat report with emissions from one or more sensors to create a shared pool of translated evidence. “Once the manual threat report has been combined with emitter data, the big picture takes shape,” Lowrance offers. “Any past projection information, when added to the new body of evidence, assists the system in developing a proposed future track.”
Gister technology removes any evidence that is irrelevant to the designed mission. To achieve this, a series of data translations and fusions occurs that combines various objective descriptors. Gister constructs an automated argument addressing the main mission requirements, including how certain factors support or impede progress toward meeting them. “The system involves framing equivalently timed task segments and matching them against external factors that provide the context for in-depth content analysis,” Lowrance offers.
Gister frame construction of battlefield operations is based on both the physical environment and enemy movement. Each frame represents a static piece of a dynamic gallery. Within a gallery, users determine which elements are positive for achieving operational goals and which are negative. By grouping the categories independently, they can examine potential scenarios out of the context of a larger, integrated picture. “Vegetation and rock formations that may pose difficulties for troop movements can be delineated more easily outside the context of a combined picture,” Lowrance explains. Each frame is examined individually. By limiting extraneous factors, more potential solutions to mission objectives can be found.
A key feature of Gister technology is the ability to determine the compatibility of independent elements within each frame. Within a gallery framework, analysis is conducted on possibilities as represented by the whole and probabilities as seen in individual frames. Bodies of evidence are grouped according to their impact on a given operation. From these groups, an argument is constructed based on the pros and cons of pairing particular elements of each frame. In situations where the paired elements are compatible in a mission-positive sense, much of the data can be discounted as not threatening to the objective. Characteristics that are negative to an operation require further analysis before a high probability of mission success can be attained.
“The construction of a structured argument is based on a combination of sensory data and manual input,” Lowrance notes. “Without one, the levels of probability within each frame will not exceed that of the entire gallery. The direction that an enemy may take is only as certain as the sum of the steps within a movement.” Through frame analysis, Gister recreates two separate views of the battlespace—one based on target content and the other based on environmental context. After each has been examined for positive and negative elements, users can determine the most probable outcome of various scenarios.
Gister’s summarization of sensory evidence eliminates extraneous details, decreasing the information in a gallery without diminishing the operational vision. “The more data that is available, the larger the positive and negative groupings become,” Lowrance explains. “With the discarding of positive elements, a more refined picture of the obstacles to mission success is formed.” Probabilities based on the compatibility of these negative elements are represented by highly compatible data providing a cleaner operational picture, and less compatible information results in a more complex picture.
A final gisting phase rounds out Gister functionality. “This procedure produces a single statement that captures the essence of a body of evidence, devoid of any uncertainty,” Lowrance relates. “With all but the necessary information for assessing the probability of a situation occurring excluded, a clean report based just on known fact is available for integration into a command and control database.” The Gister concept can be applied to military operations as easily as in solving organizational and intelligence system problems, he adds.
SRI International is currently working with the Defense Advanced Research Projects Agency on a structured evidential argumentation system that applies Gister technology to crisis warning for national security. Research continues on methods to make automated evidential reasoning more user friendly.