CRII: SaTC: RUI: Understanding and Collectively Mitigating Harms from Deepfake Imagery

CRII的这项研究采用了一种创新的方法来增强服务不足社区的网络安全和隐私。虽然生成式人工智能可以支持创造性表达并提高工人的生产力，但它也可以创建看似真实的图像、视频和音频，即所谓的深度伪造。深度造假被用来沉默、偷窃、勒索和欺诈，操纵股票价格和公众舆论，损害个人和组织的声誉。这些行为对国家安全、民间机构以及人们的个人和职业生活产生了负面影响。随着生成式人工智能技术变得越来越复杂和容易获得，任何人都可以用一张照片或音频剪辑创建一个深度假。这项研究系统地记录了深度造假如何损害服务不足的社区，并设计了新颖的、以社区为中心的解决方案，以增强对这些危害的抵御能力。该项目还支持在先进的社会计算和人工智能技术方面培训不同类型的学生，以解决紧迫的社会需求。本研究通过应用社会计算、可用安全性和基于社区的参与性研究的概念，超越了个人主义的安全和隐私方法，并使用混合方法综合了深度伪造的潜在亲社会和反社会应用，以及它们对人们个人和私人生活的影响。研究结果可以为政府和技术政策提供信息，以满足服务不足群体的需求，并设计负责任的生成人工智能工具。该研究团队正在与费城地区两个服务不足的群体，黑人社区和亚洲移民/侨民社区密切合作，建立一个可扩展的平台，以鼓励和支持社区层面的网络安全实践。该网站还将支持部署和评估研究，推进有关危害、信任、安全和隐私的基础知识。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

This CRII research is taking an innovative approach to augment underserved communities’ cybersecurity and privacy. Although generative AI can support creative expression and boost worker productivity, it also enables the creation of seemingly-real images, video, and audio —known as deepfakes. Deepfakes are being used to silence, steal, extort, and defraud, manipulate stock prices and public opinion, and harm the reputation of people and organizations. These actions have a negative impact on national security, civic institutions, and people’s personal and professional lives. As generative AI technologies become increasingly sophisticated and accessible, anyone can create a deepfake with a single photo or audio clip. This research systematically documents how deepfakes harm underserved communities and devises novel, community-centric solutions to become more resilient to these harms. The project also supports the training of diverse students in advanced social computing and AI technologies to address urgent societal need.This research transcends individualistic approaches to security and privacy by applying concepts from social computing, usable security, and community-based participatory research, and it uses a mixed-methods approach to synthesize the potentially prosocial as well as antisocial applications of deepfakes, and their impact on people’s personal and private lives. The findings could inform government and technology policy that addresses the needs of underserved groups, and the design of responsible generative AI tools. The research team is collaborating closely with two underserved groups in the Philadelphia area, Black communities and Asian immigrant/diaspora communities, to build a scalable platform to encourage and support community-level cybersecurity practices. The site will also enable deployment and evaluation studies, advancing foundational knowledge about harm, trust, and security and privacy.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.