By Gao Ge
As the saying goes, "Seeing is believing." Yet, in the age of Artificial Intelligence (AI), it's not necessarily true. With the advancement of content generation technology, particularly the refinement of deepfake technology rooted in deep learning, it becomes harder to distinguish the true from the false.
Deepfake is a coined compound word by combining "deep learning" and "fake." It refers to a method of information manipulation and replacement based on AI deep learning technology. Technically speaking, the deep learning on which deepfake relies is essentially an AI neural network capable of mimicking the human brain's ability to recognize data patterns. The most well-known AI neural network to date is the Generative Adversarial Network (GAN). This network can continuously generate and identify fake data until it becomes indistinguishable from real data. Deepfake technology can now perform functions such as face swap, facial reenactment, lip-syncing, and motion transfer, as well as image-, audio-, and text- generation. In summary, it unfolds the following characteristics.
Easy access. With numerous tutorials and open-source programs available on the Internet, anyone with a computer and basic computer skills can master deepfake technology in a short time. In addition, as the artificial neural network technology progresses rapidly and the computing power continues to evolve diversely, more people will be able to possess and utilize this capability.
High-efficiency generation. Deepfake technology, primarily relying on the artificial neural network, can operate intelligently with minimal human intervention. The network can iterate and optimize quickly and efficiently. As long as sufficient data samples are provided, they can continuously generate images, videos, or audio.
Difficult to detect and identify. Unlike its predecessors relying on relatively simplistic image, audio, and video manipulation techniques, deepfakes harness the power of AI deep learning from samples. As the GAN references more data, the authenticity of its generated content increases exponentially. Deepfakes excel in comprehensive learning by incorporating biological nuances such as the target's voice and micro-expressions. Such multi-faceted simulation is beyond the reach of previous falsification methods, presenting a daunting obstacle for detection and identification technologies to overcome.
Technology is a critical driver of shifts in battlefield tactics. Deepfake technology, with its advanced cognitive manipulation capabilities, is increasingly applicable in military contexts.
At the strategic level, deepfake technology can be utilized to fabricate and distribute fake videos and audio content, manipulating public perception, and subsequently leveraging mass media to achieve specific political and military ends. At the operational level of warfare, it can interfere decision-making processes of commanders across all ranks by disseminating false information. The Russia-Ukraine conflict stands as the inaugural testbed for the widespread application of AI in the military domain. Notably, deepfake technology has been employed to propagate disinformation, with the purpose of causing psychological disruption to adversaries. At the tactical level, the intrusion of disinformation has the potential to undermine the troops' combat morale and sow discord between superiors and subordinates, or allies.
Today, while the US and Europe are issuing warnings about the risks associated with deepfake technology, they are simultaneously harnessing its potential in developing new-generation military information tools. The US Special Operations Command (USSOCOM) has released procurement documents explicitly outlining the research and development of a new generation of Military Information Support Operations (MISO) tools, including deepfake technology. Moreover, the US Army has initiated training programs for psychological warfare personnel to utilize AI technology in creating voice cloning software. It is speculated that upon deployment on the battlefield, this technology will reshape the US military's capability to influence the target population and battlefield personnel.
In the era of intelligent warfare, the proliferation of false information generated by deepfake technology is bound to be an inevitable aspect during the whole course of modern conflict. How shall we respond to the threats posed by deepfake technology? Firstly, by enhancing research and application of deepfake detection technology through increased innovation and investment, we can provide technical support of detection technologies. Secondly, constructing a robust digital protection system is crucial. By advancing research and utilization of technologies such as deep learning and digital encryption, critical information can be safeguarded, thereby raising the threshold for the application of deepfake technology at its source. Thirdly, through public education and promotion initiatives, awareness and psychological vigilance against the dangers of deepfake technology can be raised among military personnel.
