The Joint MURI-AUSMURI in Cybersecurity Assurance for Teams of Computers and Humans will develop rigorous science for human-bot cybersecurity teams, with the goal of developing cohesive teams that are robust and anti-fragile to active human and machine learning adversaries. A significant collaborative effort, this initiative draws together eight MURI-funded researchers from Wisconsin, Carnegie Mellon, Penn State, UC San Diego and eight AUSMURI-funded researchers from the University of Melbourne, University of Newcastle, Macquarie. Our multidisciplinary team draws together expertise in cybersecurity, artificial intelligence, human-computer interaction, psychology, and decision sciences.
Automated machine learning systems aim to reduce the cognitive burden on human cyber analysts by filtering information, thus enabling greater focus on the high-level mission. However just as defenders enjoy the fruits of automation, attackers can quickly adapt to changing conditions and find flaws in automated systems. Effective coordination of human-bot teams is therefore a grand challenge for cybersecurity and the focus of this initiative.
There are three main challenges in achieving our high-level goals. First, active adversaries pose significant challenges and can interfere with the entire human-bot team. For example, an adversary might manipulate or “poison” data so that a specific threat goes unnoticed (e.g. a malware is identified as benign by a bot and thus never reaches an analyst). Second, the cybersecurity landscape is highly dynamic. New threats are discovered daily and the nature of the threats is constantly changing. This is known as distribution shift in machine learning. Rarely we see threats the likes of which have never been seen before (known as black-swan or six-sigma events). Finally, the entire cybersecurity team needs to function as a coherent trusted unit towards a common goal or mission of protecting critical assets. This means that mathematical models of robust machine learning systems and human decision making, must be developed in concert.
The project’s technical approach and research thrusts have been designed to address these technical challenges.