GenAI Mirage: The Impostor Bias and the Deepfake Detection Challenge in the Era of Artificial Illusions

TL;DR

The paper investigates how cognitive biases distort forensic and digital-forensic judgments in the era of AI-generated multimedia. It introduces Impostor Bias, a priori distrust of AI-created content, and surveys existing debiasing strategies while evaluating deepfake detection methods and their limitations. By linking bias research with technical detection traces (e.g., frequency-domain cues) and mitigation frameworks like Linear Sequential Unmasking, the work emphasizes enhancing objectivity and reliability in forensic investigations. The study highlights practical needs for advanced detectors, bias awareness, and ethical considerations to sustain trust and integrity in forensic practice amid increasingly realistic synthetic media.

Abstract

This paper examines the impact of cognitive biases on decision-making in forensics and digital forensics, exploring biases such as confirmation bias, anchoring bias, and hindsight bias. It assesses existing methods to mitigate biases and improve decision-making, introducing the novel "Impostor Bias", which arises as a systematic tendency to question the authenticity of multimedia content, such as audio, images, and videos, often assuming they are generated by AI tools. This bias goes beyond evaluators' knowledge levels, as it can lead to erroneous judgments and false accusations, undermining the reliability and credibility of forensic evidence. Impostor Bias stems from an a priori assumption rather than an objective content assessment, and its impact is expected to grow with the increasing realism of AI-generated multimedia products. The paper discusses the potential causes and consequences of Impostor Bias, suggesting strategies for prevention and counteraction. By addressing these topics, this paper aims to provide valuable insights, enhance the objectivity and validity of forensic investigations, and offer recommendations for future research and practical applications to ensure the integrity and reliability of forensic practices.

Figures (12)

• Figure 1: Eight potential sources of bias that could impact forensic decision-making.
• Figure 2: John asked Lucy for help with a license plate, reading “BC 537”, but unsure about the last two characters and the first one. Credit: fontani_cognitive_2021.
• Figure 3: Image processing could lead to produce a noticeable different image, with different face characteristics. Credit: fontani_cognitive_2021.
• Figure 4: A supposed concealed image of a face on the back side of the Shroud is revealed through advanced image processing of a photograph published in a book. The image is flipped from right to left (b). A negative image of the face that can be seen on the front side of the Shroud, processed in the same way as (a). Credit: fanti2004double.
• Figure 5: A magnified version of Figure \ref{['fig:shroud1']}b (a). A face resembling the Shroud that we discern in the top-left section of Figure \ref{['fig:shroud1']}b (as depicted on the right) (b). We can also discern another face in the bottom left section of Figure \ref{['fig:shroud1']}b. Pareidolia leads to false positives, enabling us to see faces in Figure \ref{['fig:shroud1']}b that aren't actually there. Credit: di2013pattern.