Dataset Information

Deepfake detection with and without content warnings.

ABSTRACT: The rapid advancement of 'deepfake' video technology-which uses deep learning artificial intelligence algorithms to create fake videos that look real-has given urgency to the question of how policymakers and technology companies should moderate inauthentic content. We conduct an experiment to measure people's alertness to and ability to detect a high-quality deepfake among a set of videos. First, we find that in a natural setting with no content warnings, individuals who are exposed to a deepfake video of neutral content are no more likely to detect anything out of the ordinary (32.9%) compared to a control group who viewed only authentic videos (34.1%). Second, we find that when individuals are given a warning that at least one video in a set of five is a deepfake, only 21.6% of respondents correctly identify the deepfake as the only inauthentic video, while the remainder erroneously select at least one genuine video as a deepfake.

SUBMITTER: Lewis A

PROVIDER: S-EPMC10679876 | biostudies-literature | 2023 Nov

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Deepfake detection with and without content warnings.

Lewis Andrew A Vu Patrick P Duch Raymond M RM Chowdhury Areeq A

Royal Society open science 20231127 11

The rapid advancement of 'deepfake' video technology-which uses deep learning artificial intelligence algorithms to create fake videos that look real-has given urgency to the question of how policymakers and technology companies should moderate inauthentic content. We conduct an experiment to measure people's alertness to and ability to detect a high-quality deepfake among a set of videos. First, we find that in a natural setting with no content warnings, individuals who are exposed to a deepfak ...[more]

PMID: 38026025

Similar Datasets

Project description:Content and trigger warnings give information about the content of material prior to receiving it. Different typologies of content warnings have emerged across multiple sectors, including health, social media, education and entertainment. Benefits arising from their use are contested, with recent empirical evidence from educational sectors suggesting they may raise anxiety and reinforce the centrality of trauma experience to identity, whilst benefits relate to increased individual agency in making informed decisions about engaging with content. Research is hampered by the absence of a shared inter-sectoral typology of warnings. The aims of this systematic review are to develop a typology of content warnings and to identify the contexts in which content warnings are used. The review was pre-registered (ID: CRD42020197687, URL: https://www.crd.york.ac.uk/prospero/display_record.php?ID=CRD42020197687) and used five sources: electronic databases covering multiple sectors (n = 19); table of contents from multi-sectoral journals (n = 5), traditional and social media websites (n = 53 spanning 36 countries); forward and backward citation tracking; and expert consultation (n = 15). In total, 6,254 documents were reviewed for eligibility and 136 documents from 32 countries were included. These were synthesised to develop the Narrative Experiences Online (NEON) content warning typology, which comprises 14 domains: Violence, Sex, Stigma, Disturbing content, Language, Risky behaviours, Mental health, Death, Parental guidance, Crime, Abuse, Socio-political, Flashing lights and Objects. Ten sectors were identified: Education, Audio-visual industries, Games and Apps, Media studies, Social sciences, Comic books, Social media, Music, Mental health, and Science and Technology. Presentation formats (n = 15) comprised: education materials, film, games, websites, television, books, social media, verbally, print media, apps, radio, music, research, DVD/video and policy document. The NEON content warning typology provides a framework for consistent warning use and specification of key contextual information (sector, presentation format, target audience) in future content warning research, allowing personalisation of content warnings and investigation of global sociopolitical trends over time.

Project description:ObjectiveMany countries have adopted warning labels for combustible tobacco products, yet little research exists describing tobacco warning characteristics globally and to what extent they meet the WHO Framework Convention for Tobacco Control (FCTC) Guidelines. This study evaluates characteristics of combustible tobacco warnings.DesignWe conducted a content analysis to describe the overall landscape of warnings using descriptive statistics and compared to the WHO FCTC Guidelines.SettingWe searched existing warning databases for combustible tobacco warnings from English-speaking countries. We compiled warnings meeting inclusion criteria and coded for message and image characteristics using a pre-defined codebook.Primary and secondary outcomes measuresCharacteristics of combustible tobacco warning text statements and warning images were the primary study outcomes. There were no secondary study outcomes.ResultsWe identified a total of 316 warnings from 26 countries or jurisdictions worldwide. Of these warnings, 94% included warning text and an image. Warning text statements most often described health effects to the respiratory (26%), circulatory (19%) and reproductive systems (19%). Cancer was the most frequently mentioned health topic (28%). Fewer than half of warnings included a Quitline resource (41%). Few warnings included messages about secondhand smoke (11%), addiction (6%) or cost (1%). Of warnings with images, most were in colour and showed people (88%), mostly adults (40%). More than 1 in 5 warnings with images included a smoking cue (ie, cigarette).ConclusionsWhile most tobacco warnings followed WHO FCTC guidance on effective tobacco warnings, such as communicating health risks and inclusion of images, many did not include local Quitline or cessation resources. A sizeable minority include smoking cues that could inhibit effectiveness. Full alignment with WHO FCTC guidelines will improve warnings and better achieve the WHO FCTC objectives.

Project description:IntroductionSocial interactions are a key mechanism through which health communication efforts, including pictorial cigarette pack warnings, may exert their effects. We sought to better understand social interactions elicited by pictorial cigarette pack warnings.MethodsA controlled trial randomly assigned US adult smokers (n = 2149) to have their cigarette packs labeled with pictorial or text-only warnings for 4 weeks. Smokers completed surveys during the baseline visit and each of the subsequent 4 weekly visits.ResultsSmokers with pictorial warnings on their packs had more conversations throughout the trial compared to those with text-only warnings (8.2 conversations vs 5.0, p<.01). The highest number of conversations occurred during the first week. Smokers with pictorial warnings were more likely than those with text-only warnings to discuss the health effects of smoking, whether the warnings would make them want to quit and whether the warnings would make others want to quit (all p < .05). Smokers were more likely to describe pictorial warnings as scary, gross, or depressing and gloomy during conversations than text-only warnings (all p < .05).ConclusionsPictorial warnings sparked more conversations about the warnings, the health effects of smoking, and quitting smoking than text-only warnings. These social interactions may extend the reach of pictorial warnings beyond the targeted smoker and may be one of the processes by which pictorial warnings have impact.ImplicationsHealth communication can influence behavior by changing social interactions. Our trial characterized social interactions about pictorial cigarette pack warnings with a large longitudinal sample in a real-world setting. Understanding these conversations can inform the United States and other countries as they improve existing warnings and help tobacco control policy makers and health communication theorists understand how social interactions triggered by warnings affect smoking.

Dataset Information

Deepfake detection with and without content warnings.

Publications

Deepfake detection with and without content warnings.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets