从原则到实践：利用 Indeed 负责任地扩展人工智能规模.pdf

1、 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.A I M 3 3 2 3From principles to practice:Scaling AI responsibly with IndeedMike DiamondAWSPrincipal Product LeadResponsible AILewis BakerIndeedSr.Data Science Manage

2、r&Head of Responsible AI 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.There is no escaping responsible AIDoes the generated text welcome all buyer demographic groups equally?Are property features accurately captured from inputs rather than hallucinated?Have private details abou

3、t owners or previous occupants leaked into descriptions?Do uploaded and generated images remain free of unsafe or illegal content?Condominium Description Use Case 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.There is no escaping responsible AICan the agent provide specific reco

4、mmendations while maintaining equitable service across demographic groups?Is it susceptible to accumulating unauthorized charges or exceeding budget?Does it inappropriately share personal data across supplier and payment API interactions?Is it vulnerable to manipulation attacks,such as prompt inject

5、ions designed to trigger unauthorized refunds?Shopping Agent Use Case 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.0100200300400500600Total Incidents&Hazards6-month moving averageA c c o r d i n g t o t h e O E C D A I I n c i d e n t s a n d H a z a r d s m o n i t o r Oct 202

6、5 had the highest number of reported AI incidents or hazards Number of incidents and hazards Note:An AI incident or hazard can be reported by one or more news articles covering the same event.All time total:9742J a n 2 0 2 2 O c t 2 0 2 595.02%change(year over year)October 2025:509October 2024:261 2