The Age of Synthetic Realities: Challenges and Opportunities

TL;DR

The crucial need for the development of forensic techniques capable of identifying harmful synthetic creations and distinguishing them from reality is highlighted, especially important in scenarios involving the creation and dissemination of fake news, disinformation, and misinformation.

Abstract

Synthetic realities are digital creations or augmentations that are contextually generated through the use of Artificial Intelligence (AI) methods, leveraging extensive amounts of data to construct new narratives or realities, regardless of the intent to deceive. In this paper, we delve into the concept of synthetic realities and their implications for Digital Forensics and society at large within the rapidly advancing field of AI. We highlight the crucial need for the development of forensic techniques capable of identifying harmful synthetic creations and distinguishing them from reality. This is especially important in scenarios involving the creation and dissemination of fake news, disinformation, and misinformation. Our focus extends to various forms of media, such as images, videos, audio, and text, as we examine how synthetic realities are crafted and explore approaches to detecting these malicious creations. Additionally, we shed light on the key research challenges that lie ahead in this area. This study is of paramount importance due to the rapid progress of AI generative techniques and their impact on the fundamental principles of Forensic Science.

Figures (4)

• Figure 1: Examples of image synthesis conditioning. The input modality and expected output define the type of generation task performed by the model and help express desired characteristics in the synthesized creation. Examples created with and/or reproduced from midjourney:onlineautomatic1111:onlinestablediffusiongithub:online.
• Figure 2: Classic manual image manipulation versus modern image synthesis. The image on the left (a) represents a well-known case of state-level manual image manipulation, which was misleadingly published by multiple news websites as genuine in 2008. In contrast, we generated the image on the right (b) with MidJourney by using a prompt as simple as "missile test". Cases like the former one currently represent an even greater challenge to authenticity verification.
• Figure 3: Examples of generated image inconsistencies. All images were generated with version 5.1 of the Midjourney model -- the latest one released at the writing of this article. In (a), an unnatural synthetic hand with six fingers. In (b), a synthetic wheelchair with inconsistent design; the seat orientation does not match the wheels' position. In (c), a synthetic billboard with text that makes no sense and presents aberrant letters. In (d), a synthetic paisage with cloudy skies and mountains by a lake; the highlighted cloud is not congruently reflected on the lake's surface. To generate these images, we used the following prompts: (a) "lady's hand with a ring on it", (b) "wheelchair in a hospital", (c) "outdoor sign with a religious statement on it", and (d) "realistic photo; mountains with a lake at the bottom".
• Figure 4: Human versus Machine poem writing. (a) is a public-domain poem entitled Thoughts from Myra Viola Wilds; (b) is a poem written by chatGPT when asked to write a poem entitled 'Thoughts'.