Jiawei Zhang (University of California, Berkeley Law School) & Shuang Liu (Carnegie Mellon University – School of Computer Science) have posted Deepfake & Deepfaith on SSRN. Here is the abstract:
In the age of deepfake, we need deepfaith in ourselves. Unlike most deepfake studies that catastrophize deepfakes’ impacts, this Article critically examines deepfakes’ impacts on the information marketplace dynamics and the truth-seeking objectives from a transtemporal lens. To that end, This Article evaluates the inter-informational competitive relationship between visual representations and word-based statements. In the Pre-Deepfake Age, visual representation was the most competitive information type in the truth-seeking marketplace, surpassing word-based factual statements and opinions. However, the prevalence of deepfakes is revolutionizing the already established inter-informational competitive order. From this perspective, this Article categorizes the deepfakes’ incursions into two phases: Deepfake 1.0 and Deepfake 2.0.
In Deepfake 1.0, the widespread diffusion of generative AI has significantly lowered the technical barriers to creating fabricated audiovisual content, making authentication and falsification nearly impossible. In the absence of sufficient, reliable information sources, individuals lacking necessary technical tools or literacy skills have to bear overwhelming cognitive burdens when confronted with persuasive yet deceptive content. As a result, the prevalence of deepfakes weakens the competitiveness of visual representation in the truth-seeking marketplace and distorts the longstanding inter-informational competitive hierarchy among visual representation, factual statement, and normative opinion. This unwelcome tendency forces numerous First Amendment scholars to adopt a conservative response, either by fundamentally questioning the value of false factual content protected under Alvarez, or by seeking to distinguish deepfakes from such constitutionally protected falsehoods through adaptive doctrinal strategies.
However, if we see this transformation in the long term, there will actually be a silver lining. As deepfakes become further widespread and visual literacy is sufficiently cultivated, people will be more mentally and intellectually prepared for visual uptake. They will critically analyze the presented information and compare competing sources. This process will undermine the automatic credibility of visual representations, making them far less competitive in the marketplace. This marks the entry into Deepfake 2.0, where a new market equilibrium will be built: the competitiveness of three information forms converges as people treat visual materials equivalent to factual statements, and even to subjective, unfalsifiable normative opinions. This benign “convergence” may trigger a more lenient approach to address deepfakes’ eligibility for First Amendment protection.
Our transtemporal examination of the inter-informational competitive dynamics enables us to conclude that the prevalence of deepfakes, as well as resulting “convergence,” is simply bringing us back to the Pre-Visual Age. The worst case under Deepfake 2.0 is the regression of truth certainty to the level of the Pre-Visual Age. Although such a regression is undesirable, it does not warrant excessive alarm. In preparing for Deepfake 2.0, it is worth reflecting on how our predecessors in the Pre-Visual Age navigated an age with uncertainties and continuously cultivated wisdom. In this sense, deepfaith in human resilience and intelligence remains the most powerful weapon against the challenges posed by deepfakes.