본문 바로가기 주메뉴 바로가기
국회도서관 홈으로 정보검색 소장정보 검색

목차보기

Title page 1

Contents 6

Introduction 9

1. Background 11

1.1. Generative AI 11

1.2. Data centers 12

1.3. Measuring environmental effects 13

1.4. Life cycle assessments 14

1.5. Policy environment 16

2. Generative AI Uses Significant Resources, but Environmental Effects Are Uncertain 19

2.1. Uncertain but large resource costs of training and using generative AI 19

2.2. Lack of data for generative AI infrastructure build activities 23

2.3. Insufficient information about effects from computing-infrastructure end-of-life issues 24

2.4. Unpredictable technological advancements and demand for generative AI 24

3. Generative AI Could Have Substantial Human Effects 27

3.1. Risks and challenges of generative AI development and use 27

3.2. Common industry mitigation strategies 31

3.3. Human effects of generative AI in selected applications 32

4. Policy Options for the Environmental and Human Effects of Generative AI 37

4.1. Policy options for environmental effects of generative AI 37

4.2. Policy options for human effects of generative AI 38

5. Agency and Expert Comments 41

Appendix I: Objectives, Scope, and Methodology 42

Appendix II: Expert Meeting Participants 44

Appendix III: GAO Contacts and Staff Acknowledgments 46

Tables 7

Table 1. Reported energy consumption and carbon emissions for the training of selected generative AI models 21

Table 2. Examples of common mitigation strategies to address risks and challenges of generative artificial intelligence (AI) 31

Figures 7

Figure 1. Generative artificial intelligence (AI) in relationship to other types of AI 12

Figure 2. Life cycle of generative artificial intelligence 15

Figure 3. Estimated carbon emissions and energy equivalents for a generative artificial intelligence model requiring 5,000 megawatt-hours of electricity 20

Figure 4. Estimated amounts of energy and water consumed to use generative artificial intelligence (AI) for internet searches 22

Figure 5. Selected generative artificial intelligence risks and challenges that could result in human effects 28