Confidence-Building Measures for Artificial Intelligence: Workshop proceedings 인공 지능에 대한 신뢰 구축 조치: 워크숍 절차
The Confidence-Building Measures for Artificial Intelligence (CBMs) workshop was held in December 2021 to discuss the potential of CBMs to reduce the risks of harmful AI applications. The workshop proceedings, published by OpenAI, identify six CBMs that could be applied to foundation models: Crisis hotlines: These hotlines would provide a way for people to report harmful AI applications or to seek help if they are being harmed by AI. Incident sharing: This would involve sharing information about harmful AI applications with other stakeholders, such as government agencies, researchers, and the public. Model, transparency, and system cards: These cards would provide information about the design, training, and deployment of foundation models, in order to help users understand how these models work and to identify potential risks. Content provenance and watermarks: These techniques would help to track the origin of content generated by AI models, in order to prevent the spread of harmful