You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Generative AI systems across modalities, ranging from text, image, audio, andvideo, have broad social impacts, but there exists no official standard formeans of evaluating those impacts and which impacts should be evaluated. Wemove toward a standard approach in evaluating a generative AI system for anymodality, in two overarching categories: what is able to be evaluated in a basesystem that has no predetermined application and what is able to be evaluatedin society. We describe specific social impact categories and how to approachand conduct evaluations in the base technical system, then in people andsociety. Our framework for a base system defines seven categories of socialimpact: bias, stereotypes, and representational harms; cultural values andsensitive content; disparate performance; privacy and data protection;financial costs; environmental costs; and data and content moderation laborcosts. Suggested methods for evaluation apply to all modalities and analyses ofthe limitations of existing evaluations serve as a starting point for necessaryinvestment in future evaluations. We offer five overarching categories for whatis able to be evaluated in society, each with their own subcategories:trustworthiness and autonomy; inequality, marginalization, and violence;concentration of authority; labor and creativity; and ecosystem andenvironment. Each subcategory includes recommendations for mitigating harm. Weare concurrently crafting an evaluation repository for the AI researchcommunity to contribute existing evaluations along the given categories. Thisversion will be updated following a CRAFT session at ACM FAccT 2023.
AkihikoWatanabe
changed the title
あ
Evaluating the Social Impact of Generative AI Systems in Systems and
Society, Irene Solaiman+, N/A, arXiv'23
Jun 16, 2023
URL
Affiliations
Abstract
Translation (by gpt-3.5-turbo)
Summary (by gpt-3.5-turbo)
The text was updated successfully, but these errors were encountered: