“GANs and Deepfakes”: AFCLC faculty dig into research elective at SOS

MAXWELL AIR FORCE BASE, Ala. --

From 22 July to 22 August, the Air University Integration Cell held an advanced research elective course for the Squadron Officer School on the topic of Generative Adversarial Networks (GANs) and deepfakes. Three AFCLC faculty took part in the elective as advisors to the six SOS students selected for the elective. Dr. Elizabeth Peifer, Dr. Susan Steen, and Dr. Patricia Fogarty each contributed their research interests and subject matter expertise to discussions with students.

GANs are a type of machine learning where two neural networks are given a training set of data (for example, photos of sad faces), then one network tries to generate new data of the same class (computer-generated sad faces) in an attempt to trick the other network into thinking that the new data was actually created by a human (that they are actually new photos of sad faces). Given the feedback from the second network, the first network tries to improve its computer-generated sad faces until the second network will recognize the new data as equal to original sad faces. In the process, the neural network learns the characteristics of sad faces and how to reproduce them to a higher quality standard. ,

Deepfakes are a specific use of GANs in which audio or video is created and substituted into authentic files, usually for the purpose of deception. They are similar to the many techniques that people have used over time to cut and paste content into or over other content, but the difference with deepfakes is that artificial intelligence is doing the work. ,

You might be wondering why the Culture and Language Center faculty would be interested in this SOS research elective. We look at everything from the point of view of how humans think and behave in different contexts, so it’s interesting to us how people create and manipulate technology in uniquely cultural ways. People with different views of the world enable neural networks to create new information based on patterns that may be culturally constructed. Who decides what information is given to neural networks, what information is reproduced, in what formats, and with what key characteristics? All of those things are governed by cultural context.

However, our faculty are probably most interested in how people use the outputs of GANs and deepfakes. For example, do people reproduce information with the intent of undermining foreign states or world leaders? Which states and leaders, when, where, how, and why? This is where our faculty expertise added value to the SOS research elective. Dr. Steen’s expertise concerns communication and leadership; she engaged students on the ways that people manipulate narratives and media in order to achieve military and political aims. Dr. Peifer is a historian who conducts research into extremist groups, particularly those active across Europe; she contributed to this elective by offering resources for how particular groups have used the Internet to bolster extremist ties, or make them seem more robust than they actually are. Dr. Fogarty’s interests are in Eastern Europe and in discourse analysis; she offered to the SOS students information about public and private sector organizations that are exposing deepfakes, conducting open source analysis, and carrying out research on social media usage.

During the elective, the students produced short white papers that they presented on the final day to Mr. Joseph McDade, Jr, the SES Assistant Deputy Chief of Staff in HAF/A8. Mr. McDade’s office originated the request for the research elective, asking students to examine the issue of potential defensive measures against GANs and deepfakes. Students’ work ranged from exposing the capabilities of GANs to disrupt Air Force missions, to building Airmen’s awareness of the threat of GANs, to the legal ramifications of how US law treats cyber crimes vs international cyber warfare.

Dr. Margaret Sankey, Research Coordinator at Air University and one of the coordinators of the SOS research elective, expressed gratitude for AFCLC’s faculty involvement, and said that “understanding how people react to information that challenges or supports existing beliefs is something that anthropology, communications and history offer that no amount of technological know-how can fully comprehend.”