首页|Charles Sturt University Reports Findings in Artificial Intelligence (Gender and Ethnicity Bias of Text-to-Image Generative Artificial Intelligence in Medical I maging, Part 1: Preliminary Evaluation)
Charles Sturt University Reports Findings in Artificial Intelligence (Gender and Ethnicity Bias of Text-to-Image Generative Artificial Intelligence in Medical I maging, Part 1: Preliminary Evaluation)
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
NETL
NSTL
By a News Reporter-Staff News Editor at Robotics & Machine Learning DailyNews Daily News-New research on Artificial Intelligenc e is the subject of a report. According to newsreporting out of Wagga Wagga, Au stralia, by NewsRx editors, research stated, "Generative artificialintelligence (AI) text-to-image production could reinforce or amplify gender and ethnicity b iases. Severaltext-to-image generative AI tools are used for producing images t hat represent the medical imagingprofessions."Our news journalists obtained a quote from the research from Charles Sturt Unive rsity, "White malestereotyping and masculine cultures can dissuade women and et hnically divergent people from being drawninto a profession. In March 2024, DAL L-E 3, Firefly 2, Stable Diffusion 2.1, and Midjourney 5.2 wereutilized to gene rate a series of individual and group images of medical imaging professionals: r adiologist,nuclear medicine physician, radiographer, and nuclear medicine techn ologist. Multiple iterations of imageswere generated using a variety of prompts . Collectively, 184 images were produced for evaluation of 391characters. All i mages were independently analyzed by 3 reviewers for apparent gender and skin to ne.Collectively (individual and group characters) ( = 391), 60.6% were male and 87.7% were of a light skintone. DALL-E 3 (65.6% ), Midjourney 5.2 (76.7%), and Stable Diffusion 2.1 (56.2% ) had a statisticallyhigher representation of men than Firefly 2 (42.9% ) (<0.0001). With Firefly 2, 70.3% of charac ters hadlight skin tones, which was statistically lower (<0.0001) than for Stable Diffusion 2.1 (84.8%), Midjourney5.2 (100 %), and DALL-E 3 (94.8%). Overall, image quality metri cs were average or better in 87.2% forDALL-E 3 and 86.2% for Midjourney 5.2, whereas 50.9% were inadequate or poor for Fire fly 2 and 86.0% for Stable Diffusion 2.1. Generative AI text-to-im age generation using DALL-E 3 via GPT-4 has thebest overall quality compared wi th Firefly 2, Midjourney 5.2, and Stable Diffusion 2.1."
Wagga WaggaAustraliaAustralia and Ne w ZealandArtificial IntelligenceEmerging TechnologiesHealth and MedicineMachine LearningNuclear Medicine
NETL
NSTL
