Illinois researchers receive grant to study human-machine interactions
The project emphasizes the study of communication and information acquisition and exchange between agents of different types in an adversarial environment, such as the battlefield. It is expected to have a dramatic impact on the most critical issues of inference: decision making and overall situational awareness.
“Our national security and economic health depend on our ability to provide robust, timely, and accurate responses to challenges that arise in complex networked environments where humans and machines with varying capabilities and intents interact,” said Tamer Başar, the project principal investigator who is a professor of Electrical and Computer Engineering (ECE) and the holder of a Swanlund Chair at the University. “Our work addresses that need, and is expected to have a dramatic impact on the most critical issues of inference, decision making, and overall situational awareness.”
In addition to Langbort and Başar, CSL researchers Geir Dullerud and Negar Kiyavash, and ECE Professor Rayadurgam Srikant, along with six faculty from Georgia Tech, Stanford, UC-Berkeley, and the University of Maryland are also participating. The faculty come from a variety of backgrounds, including engineering (electrical, computer, industrial, mechanical, and aerospace), computer science, and economics. The Air Force Office of Scientific Research will fund this project for five years. In total, the Department of Defense funded 32 MURI projects out of the 411 white papers originally submitted to the program back in December.
Langbort says that in a modern battlefield multiple agents in different locations make decisions. These agents can be humans or machines and they share information through networks, introducing new vulnerabilities and new ways for adversaries to be strategic.
“Your actions influence what is and isn’t observable to the other player or players. For example, in a situation on a battlefield, you either really fight the game in itself, or you can start attacking information itself, breaking into the network and starting rumors,” Langbort said. “(Players) can just fight at the network level by either physically breaking into it or, in a more covert but still dangerous way, strategically modifying the information it carries.”
The network attacks could be multifarious, including cognitive jamming, data tampering, malicious gossiping, disruption of physical links and servers, and hacking. They can also be stealthy, like a timing attack.
“The goal of this project is to understand how these strategic disruptions impact decision making and to architect the network, information flow, and decision algorithms themselves so that vulnerability to adversarial acts is minimized,” said Başar. “For that, it is important to model the assumptions that agents make about each other and the adversary, and humans and machines build such assumptions very differently.”
The team is using the framework of game theory, which is concerned with adversaries whose goals are nonaligned and who are competing with each other.
Dullerud, a professor of mechanical engineering, added that the team will examine the many levels of game theory, asking questions such as: What are the exact ways in which large numbers of both people and machines interact? What information do they share and how accurate is it? Are they telling the truth?
“The problem with this sort of interaction network is that there are too many entities, and it is not possible to model them all exactly. An interesting aspect is that if you look more closely at what appears to be a single modeling entity, it may well be a simplification of a game that is being played out on a smaller spatial or temporal scale, so one really has a game of games,” Dullerud said.
To gain a better understanding of computer and human interaction, Dullerud is developing a distributed robotics test bed using a network of hovercraft and other autonomous vehicles that can interact with both human and machine-based decision makers. “We’d like to have a cyberphysical network comprised of humans and machines to help us experimentally determine what we need to know in order to systematically predict the behavior of a potentially large-scale human-machine network.”
Dullerud believes the CSL’s multidisciplinary history will contribute to the success of the project. “Also, educationwise, this is a very innovative project that will open up many opportunities for both graduate and undergraduate research training,” he said.