The Duke Funding Alert newsletter, published every Monday, provides information on all new and updated grants and fellowships added to the database during the prior week. This listserv is restricted to members of the Duke community.
Award for Fundamental Research in Socio-Mathematics of Information and Influence
Information has become a domain of warfare that now extends well beyond compromising or disabling the data-handling platforms, and into subtler forms of content manipulation, in order to elicit specific psychological and sociological responses. The existence of systemic campaigns of disinformation is now well established, conducted or supervised by malicious state and non-state actors, using the cover of widespread social media outlets. These rely on automated software agents (“bots”), designed and engineered with increasing levels of sophistication, as well as trained human agents who can control the bot deployment, adjust the network, and refine the strategies. Modern technology greatly facilitates Internet-based programs of social hysteria propagation, propaganda, disinformation, and influence operations, and as technological capabilities improve, the threat will grow accordingly. In an environment of widespread information uncertainty, open societies lose coherence and risk disorderly devolvement, the ultimate objective of the adversaries. Relying on various, authoritative tools of data analysis and network science, the forensic evidence of this social manipulation is undeniable. Even so, this is far from an optimal situation: detection of such campaigns can be difficult and inaccurate when dealing with an adversary who can quickly adapt strategies, and finding a-posteriori evidence of malicious disinformation yields little opportunity for timely and effective counter-measure. Thus, there is a critical need for an ability to: a) detect and counteract disinformation in real-time, especially if the adversary is able to rapidly adapt, as well as; b) predict future strategies and tactics of information warfare in order to design the appropriate defenses.
Given the rapidly growing complexity of the information networks for ubiquitous access to information, as well as the pervasive deployment of malicious bots with fast expanding learning capabilities, these detection and prediction tasks can only be achieved by intelligent software, armed with highly efficient algorithms and operating on modern computing platforms. However, disinformation also relies on subtle factors, based on the study of emotional and cognitive processes inherent to the human targets. Those cannot be modeled as rational actors when under such influences, vastly increasing the uncertainty of predictive scenarios. Therefore, it is equally essential to incorporate this knowledge into the models and methods for detection and prediction of malign influence operations.
The overarching goal of this research program is to enhance and extend the understanding of the theoretical underpinnings of future information warfare, towards rapid detection, tracking and prediction of attempts at social manipulation. The problem requires the deep integration of two, currently distinct scientific fields, mathematics and social sciences. While modern mathematical methods are often and well-used in social science studies, this research program is going well beyond the state of the art and is calling for the development of a new mathematical foundation for describing, analyzing and predicting human social behavior at multiple scales and in complex and dynamic environments, thus laying the groundwork for a new field.
Key Dates: This announcement will be open to receive applications continuously until 4:00 p.m. Eastern Time (ET), August 28, 2020.
Areas of Interest
The first aspect concerns the mathematical understanding of the threat and the environment, leading to the design of efficient counter-measures. One must deal, especially, with the critical problems of malicious activity detection and inference of their sources in a very dynamic environment. The growing employment of Machine Learning (ML) methods, as well as other mathematical techniques for data analytics (e.g. topological data analysis – TDA, manifold reconstruction, etc.), provides some formidable capabilities, but is far from sufficient. For example, these tools deal with static data, not adversarial agents who can quickly adapt their strategies, effectively changing the rules of the game. The behavioral dynamics also evolve over a complex multi-layer network that includes cyber, media and a multitude of social dimensions. The true nature and scaling properties of these real-life networks may still be elusive, possibly disrupting an accurate interpretation of measurements, and these networks will evolve rapidly as technology and social conditions change. These are fundamental and unresolved problems for which current techniques are insufficient.
The social dimension is the second aspect of the problem. It is already well appreciated how different forms of communication influence individual and group understandings of knowledge and behavior, but the acceleration of cyber influence by sophisticated software agents could have significant implications for social trust and how people live, learn, and communicate. Questions range up to the fundamental issues of how manipulating information at greater speeds and with greater specificity and customizability affects the social relations on which society relies. With the advent of 5G technology and beyond, people may be instantaneously exposed to a deluge of information, from the geo-political to the trivial. How does this accelerated information and network complexity impact social relations across micro, meso, and macro scales? Are new ways of manipulating opinion, even subtler than before, made accessible, and could we detect them? Who then become the principal agents of manipulation, whether complicit or unwitting?
While relying on fundamental and exploratory studies in their respective fields, these two directions must eventually be integrated, as the mathematical models of the network, their agents, and the social behavior, must be based on realistic models at multiple scales of aggregation. The comprehensive basic research being sought-after in this opportunity should provide the basis for a unique ability to detect and predict evolving malign social influence, in realistic and ever-changing conditions.
The fundamental science behind the objective of this topic covers multiple, coupled areas, thus requiring a combination of expertise, for example: computer science and machine learning, mathematics, cognitive psychology and sociology, network theory and/or game theory.
Some specific research topics to be addressed in this undertaking may include, but are not limited to, the following:
1) Carefully designed mathematical abstractions based on behavioral science for modeling the agent’s psychological and social variables, e.g.: emotional and cognitive states, human intent and belief, and group dynamics. These models should include approaches to multi- scale clustering for accurate comprehension and modeling of aggregate behavior, e.g. individual – group – nation.
2) Game-theoretical and Machine Learning concepts, e.g. multi-agent reinforcement learning (RL) or distributional RL, as well as other innovative ideas that can consider a hybrid distribution of irrational and rational agents, including artificial ones (e.g. bots).
3) Efficient mathematical methods and algorithms to detect malicious intent and learn agent behavior and objectives from limited and noisy observations.
4) Concepts and methods for strategy optimization (inverse design), which may include counter-messaging, network-based intervention, or other means.
This opportunity is limited to accredited U.S. institutions of higher education (IHE) with doctoral degree-granting programs, having a campus located in the U.S., or its territories and possessions, and University Affiliated Research Centers (UARCs). DoD laboratories and other institutions and Government-sponsored institutions, such as Federally Funded Research and Development Centers (FFRDCs), are not eligible to apply.
The PI may submit only one (1) application in response to this FOA. There is no limit on the number of applications by the institution. A co-PI may also be a PI or co-PI on another proposal.