Representing Humans Proves Problematic
Developers strive to create accurate portrayals of behavior, believe breakthroughs loom.
Human behavior modeling can result in digital characters such as the ones in simulations and video games or in charts, graphs and reports that predict and explain actions. Accurately modeling humans is a difficult task because of the variables involved.
Advancements in human modeling soon could improve how military troops train and prepare for missions as well as enhance leaders’ abilities to predict how foreign cultures will react to their actions. Scientists and researchers from the military, private industry and academia are examining how to depict accurately human reactions from a variety of cultures, how to store this information in a database to make it accessible for new developments and how to keep costs and time lines reasonable. Many experts in the human modeling field expect major enhancements and new uses in the next few years.
Human behavior modeling (HBM) is the formal characterization of some aspect of human reasoning and/or performance. The results range from the digital simulations of characters in a video game to charts and graphs that predict behavior.
The HBM field has produced many training and planning tools for the
Creating a high-fidelity model currently costs several million dollars. The process is more involved and difficult than creating representations of weapons, vehicles and other objects because of the variations and complexities inherent in humans. Models of weapon systems, for example, will have minor variations and a very tightly grouped set of outcomes. Hawkins explains that modeling humans is more challenging because of the potentially inordinate complexity of the factors that drive behavior and because developers have yet to come to a decision on how to represent the behavior.
Developers also must create models that represent the flexibility and adaptability of human reasoning and behavior. According to Dr. Randolph M. Jones, senior scientist and senior adviser on strategy and technology for Soar Technology Incorporated,
Despite the difficulties and expense, Hawkins and Jones believe that in the long term, models could provide a cost savings to the government. “I started out with the observation that if we had good models, we could reduce the personnel costs associated with simulations for training and analysis,” Hawkins says. By using models, the military could save the costs involved to hire role players for exercises and other activities. In addition to paying for the role players’ time during the training, the military is required to budget for the expense of training the actors’ for their parts. By replacing them with software, the government saves personnel costs and receives a boost in flexibility. The problem Hawkins discovered is that the funds recovered in personnel costs are offset by the price of creating the models. “They’re not used as much as we would like to use them because no one wants to pay the upfront costs of creating them,” he states.
One way developers can lower the cost of HBM is by taking advantage of what others already have developed. “One of the real problems in this area has been reuse,” Hawkins says. Historically, HBM projects have included poor documentation, and many of the original code writers have moved on to new projects and places. Researchers have found it easier to start from scratch than to try to alter existing models for their needs. An effort is underway to create large-scale databases of models and their components so that developers can evaluate which models are available and how they could be applied with modifications to a new need. Personnel at the
Jones says that private industry also is pushing for databases. However, he believes that much of the information developers need will not fit well into that form. Instead, he says the industry needs a tool that has database elements and elements to make models in reusable forms. Research into such a database-type product is ongoing.
Just as Hawkins, Jones sees HBM as an economic benefit with models potentially assisting or replacing human experts, who are generally expensive to train, difficult to replace and in high demand. “The more we can formalize and emulate the knowledge that these experts use, the cheaper it will be to exploit that knowledge in recurring applications,” Jones shares. “This becomes particularly important as technology and science accelerate. It is very difficult for humans to keep up with all the information and innovations that are out there, and formalized human behavior models will prove increasingly important to manage all of this information.”
Another important trend in HBM involves creating models that capture cultural, ethnic and socioeconomic variations of non-Western cultures. Hawkins explains that one of the problems
Jean MacMillan, chief scientist at Aptima Incorporated,
MacMillan and Jones both say that most models are based on an aggregate of humans rather than on individuals, whether the result is a single-person model or a model of a group or team. MacMillan explains that individual models often encompass too much information to make their creation computationally realistic. In her work, she develops models of humans working together in teams and groups. If her tasking were to model how the U.S. Army would react to a situation, she would not represent every soldier. Instead, she would determine the important clusters of troops and try to represent those. When attempting to model a large group, the elements in the model are divided into higher level groups of people. One challenge to modeling humans is that universal models will not provide accurate results. The variability in human behavior and lack of theory prevent researchers from making the precise predictions possible for vehicles and weapons.
Social interaction modeling is becoming a hot area of research in the HBM field. Researchers want to create models that have better social skills as opposed to models built to perform a stand-alone task. Models built with social reaction skills could perform tasks such as asking others for information or assigning work to teammates, creating a more realistic dynamic. MacMillan believes the transition from individual models to social models is a cusp area for HBM that will experience near-term progression.
Many HBM experts expect major advancements in the field in the next five to 10 years. Hawkins says that his organization will invest several million dollars per year in new and advanced research, including work at education institutions that will automate the high-fidelity process. In that process, a U.S. Navy pilot could sit in an F-18 simulator running air-to-ground combat mission scenarios. The simulator would observe the performance of that person and conduct a cognitive task analysis based on the pilot’s interaction with the environment, and a to-be-developed tool would produce the code automatically.
In addition, Hawkins is interested in models with self-explanation capabilities. This feature would allow trainees to ask questions of their simulated allies or adversaries to understand better why an event occurred. This also would help developers validate their work because if the model’s explanations were nonsensical, they would know the model was invalid. Validation/verification is one of the three main cost drivers in HBM. The other two, according to Hawkins, are personnel to determine what needs to be part of the model and personnel to write the code to create the model.
Hawkins supports programs that reduce the cost of building models. He says the question now is whether the models can be scaled up from simple tasks to more complex ones. He believes that if developers can achieve the scaling, it will be a breakthrough in terms of the military being open to using the models more often and in a widespread fashion.
Also in the next 10 years, Hawkins expects cognitive neuroscience to play a large role in HBM. Scientists use techniques such as magnetic resonance imaging (MRI) to see which parts of the brain react and in what sequence when an individual performs a task. HBM developers can use that information to make better models or to evaluate current models. Hawkins explains that a valuable synergy is taking place between neuroscience and HBM theories and models.
Jones believes that for models to represent humans more accurately, different computational processing is required. Computer programs are written for sequential control; however, humans can reason in different patterns and can be interrupted, focus on the interruption, then return to the original topic. Humans can take information they have learned in various situations and apply it as appropriate, unlike computational processes.
Jones recommends that builders of human behavior models focus on their requirements and what they want the models to do specifically. Models always can be improved and therefore never will be complete, so developers have to manage their personal expectations. He adds that developers are a long way from creating a model that can fully replicate all the behaviors of a human.