Is Seeing Truly Believing? Rise of Deepfakes & The Future of Trust
At one time or another, you have probably heard or used the saying “seeing is believing,” which is widely used to express the notion that sight is the most reliable sense, and that people are more likely to believe what they see for themselves. Present-day technological advancements, particularly in the fields of Artificial Intelligence and Machine Learning (AI/ML), have led to the exponential growth and proliferation of deepfakes – a threat that continues to warp the fabric of reality and what we can believe to be true, and one that is difficult to ignore.
Simply put, deepfakes (a portmanteau of “deep learning” and “fake”) are synthetic media – audio-visual, visual only, or audio only – generated using deep learning approaches characteristic of the machine learning subset of artificial intelligence. According to Jafar et al. (2020), anyone with basic technical skills can create and share deepfakes on the internet. This capability is further accelerated by the significantly lower cost of computing and the availability of smaller but powerful devices, including smartphones and laptops.
To fully appreciate how deepfakes are shaping how we process and gauge the credibility of digital information, we will explore some associated risks, the state of detection and awareness, and regulatory frameworks.
Threats Posed by Deepfakes
It is worth noting that deepfakes have legitimate and practical applications in areas such as entertainment, digital marketing, data privacy, education, and health. In one instance, Buo (2020) notes that the World Health Organisation (WHO) leveraged an AI solution to help individuals overcome tobacco addiction and create a plan to quit smoking. While these uses are actually legal and beneficial to the public, we have already seen deepfakes being abused by malicious actors for a variety of nefarious reasons that threaten judicial systems, political environments, and businesses.
With continued digital transformation, electronic evidence continues to play a critical role in establishing facts during cybercrime investigations; therefore, its integrity must be maintained for it to be credible in any court of law. According to Buo (2020), evidence tampering or fabrication threatens judicial processes. In 2020, for instance, a woman in the United Kingdom (UK) presented a falsified audio recording as evidence in a custody case in an attempt to make the father sound like he was threatening her to paint him as violent. Further investigation revealed that she used online deepfake tutorials to learn how to create the fake recording.
Similarly, increased internet adoption and virtually unrestricted access to information make it difficult to ignore the growing threat of misinformation, which can be weaponised using targeted disinformation and malinformation for divisive politics that can destabilise political environments through dissemination of propaganda. Such misinformation can take a variety of forms, including fake news, conspiracy theories, and clickbait. Santos-d’Amorim & Miranda (2021) note that misinformation – manipulated information – is harmful to society, whether disseminated consciously with intention to deceive or otherwise. In the United States (US), for instance, there have already been cases where deepfakes have been used to make high-ranking politicians say and do things they never actually did. In one case, Donald Trump shared a fake video of former Speaker of the US House of Representatives, Nancy Pelosi, that made her appear drunk to sway public perception of his critic. The video was shared over 2.5 million times within a short time and caused public outrage, indicating the reach that the post had.
Businesses are also not spared as they continue to face new and unfamiliar fraud risks that lead to heavy financial losses fueled by deepfake-related social engineering attacks. Emerging technologies have completely changed how we think about elaborate schemes such as Business Email Compromise (BEC) attacks, where senior company officials and executives are impersonated by scammers motivated by financial gain or unauthorised access to sensitive information that can be sold on the dark web or used as leverage to extort an extravagant ransom from affected businesses. In 2020, for instance, deepfake technology was used in BEC where the CEO of a firm based in the United Arab Emirates (UAE) had his voice cloned in an impersonation attack that led to the fraudulent transfer of USD 35 million. More recently, in 2024, a junior employee at a multinational firm in Hong Kong was tricked into transferring a total of $200 million Hong Kong dollars (about USD 25.6 million) after fraudsters used deepfake technology to impersonate the firm’s Chief Financial Officer (CFO) and other employees in a video call using publicly available video and other footage.
Advancements in Detection and Awareness
Years ago, when deepfake technology was in its infancy, it was possible to quickly identify fake videos and images visually. As the technology continues to mature, this has become extremely difficult, and technology-based solutions are now the favoured option to ascertain the integrity of images, videos, and audio files. In recent years, numerous researchers have proposed various methods to facilitate the detection of deepfakes, with a lot of solutions essentially using AI in a “fight fire with fire” approach. Even traditional forensic tools have widely adopted various AI-driven models to try to give investigators the edge that they need to distinguish deepfakes from genuine media. Organisations such as Reality Defender, for instance, have dedicated specialised research resources to the detection of deepfakes, and their results are promising.
Unfortunately, even though there is a lot of advancement and promising research around the detection of deepfakes, a lot of the solutions are too technical and often require more expertise than the average person has. Additionally, the awareness of deepfakes and the harm that can arise from their abuse is still not widespread among the public, even though researchers have argued that awareness of the technology could serve as a proactive countermeasure and mitigate some of the risks posed by deepfakes.
Regulating Artificial Intelligence
Ahmed et al. (2021) note that besides widespread unawareness of deepfake technology, a greater concern is that developing countries have little to nonexistent countermeasures against deepfakes, in contrast to developed countries. In many countries around the world, for instance, there is a lack of proactive detection strategies that leverage AI and ML, as well as regulatory frameworks that govern the acceptable use of AI. However, there have been some positive regulatory advancements in the field of artificial intelligence, as is evident from the European Union (EU) recently enacting the EU AI Act (2024). With a lot of recent focus being on AI, this new regulation is set to shape Europe’s digital future, as it sets the tone for the acceptable use of AI and the mitigation of risks arising from the adoption of the technology. Similar to the EU’s General Data Protection Regulation (GDPR), it would also not be a surprise for the EU AI Act to serve as a blueprint for the rest of the world as a means to manage the harmful impact of deepfake technology.
So, What Now?
Having explored some legitimate applications of deepfake technology, it is clear that deepfakes are not entirely negative. However, deepfakes can and have been abused on numerous occasions. Such abuse will likely continue well into the future in the face of exponential technological advancement, inevitably shaping how we assess the credibility of information we interact with daily. Further, it is critical to promote general awareness of deepfake technology and its associated risks, coupled with ongoing research in deepfake detection. Ultimately, regulatory frameworks will become increasingly essential in ensuring AI technology is not abused.
References
Ahmed, M. F. Bin, Miah, M. S. U., Bhowmik, A., & Sulaiman, J. B. (2021, July 4). Awareness to Deepfake: A resistance mechanism to Deepfake. 2021 International Congress of Advanced Technology and Engineering, ICOTEN 2021. https://doi.org/10.1109/ICOTEN52080.2021.9493549
Buo, S. A. (2020). The Emerging Threats of Deepfake Attacks and Countermeasures. https://doi.org/10.13140/RG.2.2.23089.81762
Jafar, M. T., Ababneh, M., Al-Zoube, M., & Elhassan, A. (2020). Digital Forensics and Analysis of Deepfake Videos. 11th International Conference on Information and Communication Systems [ICICS], 53–58. https://doi.org/10.1109/ICICS49469.2020.239493
Santos-d’Amorim, K., & Miranda, M. (2021). Misinformation, disinformation, and malinformation: clarifying the definitions and examples in disinfodemic times. Encontros Bibli Revista Eletrônica de Biblioteconomia e Ciência Da Informação. https://doi.org/10.5007/1518-2924.2021.e76900
WHO. (n.d.). Using AI to quit tobacco. Retrieved March 15, 2022, from https://www.who.int/news-room/spotlight/using-ai-to-quit-tobacco