Deepfakes, misinformation and disinformation and authenticity infrastructure responses: Impacts on frontline witnessing, distant witnessing, and civic journalism

Abstract

Frontline witnessing and civic journalism are impacted by the rhetoric and the reality of misinformation and disinformation. This essay highlights key insights from activities of the human rights and civic journalism network WITNESS, as they seek to prepare for new forms of media manipulation, such as deepfakes, and to ensure that an emergent “authenticity infrastructure” is in place to respond to global needs for reliable information without creating additional harms. Based on global consultations on perceived threats and prioritized solutions, their efforts are primarily targeted towards synthetic media and deepfakes, which not only facilitate audiovisual falsification (including non-consensual sexual images) but also, by being embedded in societal dynamics of surveillance and civil society suppression, they challenge real footage and so undermine the credibility of civic media and frontline witnessing (also known as “liar’s dividend”). They do this within a global context where journalists and some distant witness investigators self-identify as lacking relevant skills and capacity, and face inequity in access to detection technologies. Within this context, “authenticity infrastructure” tracks media provenance, integrity, and manipulation from camera to edit to distribution, and so comes to provide “verification subsidies” that enable distant witnesses to properly interpret eye-witness footage. This “authenticity infrastructure” and related tools are rapidly moving from niche to mainstream in the form of initiatives the Content Authenticity Initiative and Coalition for Content Authenticity and Provenance, raising key questions about who participates in the production and dissemination of audiovisual information, under what circumstances and to which effect for whom. Provenance risks being weaponized unless key concerns are integrated into infrastructure proposals and implementation. Data may be used against vulnerable witnesses, or the absence of a trail, for legitimate privacy and technological access reasons, used to undermine credibility. Regulatory and extra-legal co-option are also a fear as securitized “fake news” laws proliferate. The investigation of both phenomena, deepfakes and emergent authenticity infrastructure(s), this paper argues, is important as it highlights the risks related both to the “information disorder” of deepfakes as they challenge the credibility and safety of frontline witnesses and to responses to such “disorder,” as they risk worsening inequities in access to tools for mitigation or increasing exposure to harms from technology infrastructure.

Engaging with deepfakes: a meta-synthesis from the perspective of social shaping of technology theory

“…While earlier works had highlighted the negative effects of deepfakes, this was one of the first to include a first-person description of victims' experiences. Amidst rising concerns over a post-truth era ushered in by deepfakes, Gregory (2022) highlights the efforts of WITNESS, a human rights non-profit organization in mitigating misinformation, restoring truth and curbing the malicious use of deepfakes, including the role of authenticity infrastructure in the process through a case study drawing upon the author's own experience and that of the author's colleagues in the organization. As deepfakes' technological prowess continued to expand, Ghazi-Tehrani and Pontell (2021) explored the potential for deepfake-led phishing attacks, which may be extremely challenging to decipher.…”

“…These anti-deepfake arguments also emphasize the gendered nature of the narrative emerging from the process and the violence connected to it (Eelmaa, 2022; Flynn et al. , 2021; Gregory, 2022; Newton and Stanfill, 2020; Widder et al. , 2022; Winter and Salter, 2020) including cybercrime (Ali et al.…”

“…, 2021; Gamage et al. , 2022; Ghazi-Tehrani and Pontell, 2021; Gregory, 2022; Sabrina, 2022; Widder et al. , 2022) and explicit deepfakes (Eelmaa, 2022; Fikse, 2018; Flynn et al.…”

The MeVer DeepFake Detection Service: Lessons Learnt from Developing and Deploying in the Wild

“…Another interesting approach to countering the challenge of digital media manipulation is the Content Authenticity Initiative (CAI) [18], which proposes a toolset to track the origin and manipulation history of media via an embedded Content Record. It tracks, among other things when a specific media file was produced and by whom, what editing was performed and with what tools as well as the original file before any manipulations occurred.…”

“…As with every software project, it is important to have a clear versioning scheme. For the presented service, we have decided to use an adaptation of the Semantic Versioning 2 scheme 18 . In particular, we follow the 𝑥.𝑦.𝑧 scheme where:…”