When: Friday, November 6, 1:00 PM
Where: Tyler 106<\/p>\n\n\n\n
Abstract<\/strong> Bio<\/strong> When: Friday, November 6, 1:00 PMWhere: Tyler 106 AbstractDespite the global deployment of text-to-image (T2I) models, their safety frameworks are largely calibrated to a Western-centric default, creating significant vulnerabilities for the rest of the world. To embrace cultural pluralism and bring historically under-represented perspectives in T2I safety, we conduct localised community-centered red teaming studies in […]<\/p>\n","protected":false},"author":3300,"featured_media":19282,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":"","_links_to":"","_links_to_target":""},"categories":[34,85],"tags":[],"class_list":["post-19278","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-news","category-seminars"],"acf":[],"_links":{"self":[{"href":"https:\/\/web.uri.edu\/cs\/wp-json\/wp\/v2\/posts\/19278","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/web.uri.edu\/cs\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/web.uri.edu\/cs\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/web.uri.edu\/cs\/wp-json\/wp\/v2\/users\/3300"}],"replies":[{"embeddable":true,"href":"https:\/\/web.uri.edu\/cs\/wp-json\/wp\/v2\/comments?post=19278"}],"version-history":[{"count":3,"href":"https:\/\/web.uri.edu\/cs\/wp-json\/wp\/v2\/posts\/19278\/revisions"}],"predecessor-version":[{"id":19283,"href":"https:\/\/web.uri.edu\/cs\/wp-json\/wp\/v2\/posts\/19278\/revisions\/19283"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/web.uri.edu\/cs\/wp-json\/wp\/v2\/media\/19282"}],"wp:attachment":[{"href":"https:\/\/web.uri.edu\/cs\/wp-json\/wp\/v2\/media?parent=19278"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/web.uri.edu\/cs\/wp-json\/wp\/v2\/categories?post=19278"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/web.uri.edu\/cs\/wp-json\/wp\/v2\/tags?post=19278"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
Despite the global deployment of text-to-image (T2I) models, their safety frameworks are largely calibrated to a Western-centric default, creating significant vulnerabilities for the rest of the world. To embrace cultural pluralism and bring historically under-represented perspectives in T2I safety, we conduct localised community-centered red teaming studies in the Global South. Our two-fold approach prioritizes localization and participation, by focusing on secondary urban centers in these regions, and conducting community engagement and training workshops to contextualize local norms. As a result, we present PLACES, a dataset comprising over 26,000 examples of T2I model failures collected in partnership with universities in Ghana, Nigeria, and two regions of India (Karnataka and Punjab). Analysis of prompts collected reveals a wide-ranging diversity in socio-cultural and linguistic attributes, when compared to existing geography-agnostic crowdsourced red-teaming data. We observe unique adversarial patterns enabled by local cultural and linguistic nuances, and distinct clusters within region around specific themes, such as religion in India. Moreover, we uncover structural contextual gaps in existing safety frameworks by identifying novel harms showing normative dissonance (e.g., violating religious norms, ignoring local customs, and ominous symbolism). This work argues that expanding T2I safety requires moving beyond mere scale to incorporate deeply localised, participatory methodologies for data collection and contextualization. Content warning: This paper includes examples containing potentially harmful or offensive content.<\/p>\n\n\n\n
Charvi Rastogi is a Senior Research Scientist at Google DeepMind, where her work focuses on evaluating AI and its impact on complex social systems. Under the broad umbrella of sociotechnical AI evaluation, her work has focused on two directions: pluralism in AI evals and human-AI collaboration. Towards plurality, she has\u00a0led capacity building efforts to align AI performance with diverse perspectives by designing online challenges and participatory methods for collecting varied feedback\u00a0on multimodal AI systems. The datasets are shared publicly to support the representation of pluralistic viewpoints on AI safety, highlighting voices from underrepresented cultures, geographies, and demographics.
Towards examining the harms of AI-assistance, her work uses careful experiment design to uncover the factors at play and the outcomes of AI assistance on people’s decisions. She is interested in understanding the effects of AI assistance on human cognition and expertise, knowledge sharing, and more broadly impacts on labour and information ecosystems.
Trained in the field of statistics, machine learning and human-computer interaction, she received a PhD from Carnegie Mellon University. Her dissertation focused on the\u00a0role of humans in the Machine Learning pipeline from data generation at the time of training to expert collaboration with deployed systems. Her interdisciplinary research brings together human sciences such as psychology, sociology, and cognitive science with Machine Learning research to study technological\u00a0development and deployment in complex human-driven\u00a0systems.<\/p>\n","protected":false},"excerpt":{"rendered":"