Linguistic Geometry: From fighting wars to computing them
What if conventional armed forces were equipped with a tool that changed the way wars were fought forever? What if every military unit had something like an animated X-ray into the future that showed all enemy activities and how best to protect against them? And what if these predictions were updated in real time? Take it a step further and imagine a future without missiles or nuclear warheads and with limited conventional weapons. What if the outcome of wars was decided without actual fighting but by computers instead? According to Department of Computer Science and Engineering Professor Boris Stilman and his theory of linguistic geometry (LG), it’s not a question of what if; it’s a question of when.
“Currently, little by little, the U.S. Army is adopting our LG software to global intelligence systems in stationary and mobile command posts around the world, to command and control systems inside thousands of infantry assault vehicles, and even to soldiers’ handhelds,” Stilman says. “In a couple of years, or even sooner, this visionary software will start saving lives of American soldiers and, maybe, start changing the course of wars around the world.”
LG is a type of game theory discovered by Stilman that allows people to solve classes of adversarial games of practical scale and complexity. It is ideally suited for problems that can be represented as abstract board games, for example, military decision aids, intelligent control of unmanned vehicles, simulation-based acquisition, high-level sensor fusion, cyberwar, robotic manufacturing and more. The advantage of LG is that it provides extraordinarily fast and scalable algorithms to find the best strategies for concurrent multi-agent systems. Unlike other gaming approaches, the LG algorithms permit modeling a truly intelligent enemy. LG is applicable to the non-zero-sum games and to the games with incomplete information, for example, imperfect sensors, weather and enemy deception.
Stilman’s research on new game theory started in 1972 in Moscow. For 16 years he was involved in the advanced research project PIONEER led by former world chess champion Mikhail Botvinnik and funded by the (former) U.S.S.R. State Committee for Science and Technology. The goal of the project was to discover and mathematically formalize the methodology used by the most advanced chess experts in solving chess problems; in other words, to mathematically replicate human thinking. Over the course of the project, Stilman developed the theoretical foundations of a new approach that showed its power far beyond the initial chess problem. This became the basis for the development of LG.
In 1991, Stilman joined the Department of Computer Science and Engineering at the University of Colorado Denver, and in 1999, he founded STILMAN Advanced Strategies to lead the development of LG applications and to test and transition them for use in society. STILMAN was founded with the encouragement and approval of the CU Denver administration.
“Our work with STILMAN was openly encouraged,” says Stilman. “STILMAN founders have always believed that a collaborative relationship with the university is mutually beneficial.”
Putting LG to the test
The Defense Advanced Research Projects Agency (DARPA) is the primary research agency at the U.S. Department of Defense and is one of the main defense research agencies in the world. It funds the development of technologies that may lead to revolutionary improvements in warfighting and to technology in general. In 1999, a series of LG-focused proposals to DARPA yielded a success for Stilman. As part of the team led by the Rockwell Science Center, he received a grant to develop the LG-based command and control system for the Joint Force Air Component Commander project of the U.S. Air Force. It was through this project that Stilman and his team—including members from CU Denver, several other universities and STILMAN—developed the first full-scale software prototype of the LG defense application.
However, significant progress in the development and testing of LG applications and technology transfer didn’t happen until 2004 when STILMAN was awarded the DARPA real-time adversarial intelligence and decision-making (RAID) project, a highly ambitious project in artificial intelligence aimed at developing automated tools to perform predictive analysis of enemy behavior, actions and intentions. It was time to apply LG technology to complex military operations and to test its advantages.
For the RAID project, DARPA chose one of the most difficult types of operations—Military Operations in Urban Terrain—similar to those conducted by the U.S. Army in Iraq. Though the smallest entity on the team, STILMAN was responsible for the key item: an LG-based “brain” behind the software oracle RAID that predicts the future for human adversarial teams Blue and Red. As part of such prediction, this oracle estimates enemy courses of action and suggests the best responses for the Blue team against the actions of the Red team insurgents in real time.
Over the course of the project, DARPA and the U.S. Army tested RAID software in six experiments, some of which lasted more than a month. In each experiment—following recommendations provided by RAID—the Blue team, simulating the U.S. Army, fought the Red team of insurgents. Both teams used the U.S. Army simulation package OneSAF. The two teams were housed in different rooms, and the Red team didn’t know whether it was fighting with a RAID-assisted Blue commander or a human-assisted Blue commander. In all the experiments the RAID-assisted Blue team outperformed the human-assisted Blue team and consistently defeated the Red team.
“After each simulated fight, DARPA requested the Red commander to answer the question, ‘With whom have you just fought, humans or RAID?’” says Stilman. “In 44 percent of the cases, the Red commander was wrong. In a sense, RAID successfully passed the informal Turing Test of whether it is true artificial intelligence.” He cautions, however, that like any technology, weapon or tool, RAID must be applied properly. “Military advisors to DARPA see great opportunities in RAID, but also warn about the need for appropriate use.”
After 30 projects over the last 15 years, Stilman believes a transition to the LG technology is finally happening. A growing number of applications of LG have passed comprehensive testing and are currently being applied in real-world command and control systems in the United States.
Historically, LG was developed by generalizing experiences of advanced chess players. Fifteen years of successfully applying LG to a highly diverse set of modern military operations has led Stilman and his team of researchers to believe that LG is something more fundamental than yet another mathematical model of efficient wargaming.
“I suggested that LG is a mathematical model of human thinking about armed conflict resolution, a warfighting model at the level of superintelligence,” he says. “To explain its chess-related heritage, we should recall that the game of chess was originally invented 1,500 years ago as a gaming model of ancient wars. To formally prove this hypothesis we should have demonstrated the power of LG on ancient wars that happened before the game of chess had yet to be invented. So far, we demonstrated this theoretically on major battles of Alexander the Great, Hannibal, and Julius Caesar.”
So, perhaps in the not-so-distant future, the U.S. armed forces will be able to better predict the strategies of their enemies, and modern society will have access to “what-if” analysis of historic battles, all because of the mathematical models contained in LG, and a scientist’s initial interest in strategies used in the game of chess.
Leave a Reply