241205AI safety initiative in Japan

Transcript

1. SC 42 – Artificial Intelligence AI safety Initiative in Japan AISIJapan AI Safety Institute
2. SC 42 – Artificial Intelligence Overview of AISI AISI Japan AI Safety Institute
3. SC 42 – Artificial Intelligence Innovation Strategy 2024 ① AI innovation and the acceleration of innovation through AI • Strengthening R&D capabilities (including data supply) • Acceleration of the use of AI • Enhancing AI infrastructure • Human resource development and recruitment ② Realize the AI Safety • Governance and rules • AI safety • Prevention of mis/dis information • Intellectual property rights ③ International cooperation/collaboration AISIJapan AI Safety Institute
4. SC 42 – Artificial Intelligence AISI(AI Safety Institute)  J-AISI is an organization formed with the cooperation of 10 relevant ministries and 5 related organizations. AISI Executive director IPA Deputy executive director Council of AISI AISI Steering Committee Secretariat Strategy & planning Team Technology Team Framework Team Standards Team Security Team Thematic Subcommittee Partnership Project Government policy for AI safety assurance. Report on business policies, plans, results, etc. Installed as needed Cabinet Office Relevant Ministries and Agencies: • Cabinet Office (Secretariat of Science, Technology and Innovation Policy) • National Security Secretariat • National Center of Incident readiness and Strategy for Cybersecurity • National Police Agency • Digital Agency • Ministry of Internal Affairs and Communications • Ministry of Foreign Affairs • Ministry of Education, Culture, Sports, Science and Technology • Ministry of Economy, Trade and Industry • Ministry of Defense Related organizations: • IT Promotion Agency, Japan (IPA) • National Institute of Information and Communications Technology • RIKEN • National Institute of Informatics • National Institute of Advanced Industrial Science and Technology AISIJapan AI Safety Institute
5. SC 42 – Artificial Intelligence Role and Scope of AISI • AISI supports the government by conducting surveys on AI safety, examining evaluation methods, and creating standards. • As a hub for AI safety in Japan, AISI will consolidate the latest information in industry and academia, and promote collaboration among related companies and organizations. • Collaborate with AI safety-related organizations. • AISI is not an R&D organization. • Scope • Set the scope flexibly in the following AI related issues, while considering global trends. • Social Impact • governance • AI System • contents • data AISIJapan AI Safety Institute
6. SC 42 – Artificial Intelligence Activities AISI Japan AI Safety Institute
7. • Clarify what each stakeholder should address in the flow of AI utilization AI Guidelines for Business AISIJapan AI Safety Institute Ministry of Internal Affairs and Communications(MIC) and Ministry of Economy, Trade and Industry(METI), 2024-04 Industry-specific guidelines - Financial sector - others
8. Japan-U.S. Crosswalk iso-iec-42001 AI RMF Reference FDIS 23894. OECD/EU/EO13960 Singapore AI Verify BSA Framework AI Taxonomy of trustworthiness for AI AI GfB U.S. point of view Japanese Perspective • Confirmation of the interrelationship between the U.S. NIST AI Risk Management Framework (RMF) and the Japanese AI Guidelines for Business (GfB) AISIJapan AI Safety Institute
9. • Crosswalk 1 [Terminology] (Released on April 30th) https://aisi.go.jp/assets/pdf/AISI_Crosswalk1_RMF_GfB_ver1.0.pdf • Crosswalk 2 [Contents] (Released on September 18th) https://aisi.go.jp/assets/pdf/AISI_Crosswalk2_RMF_GfB_ver1.0.pdf Results of Japan-U.S. Crosswalk AISIJapan AI Safety Institute
10. • The significance and use of the Evaluation Perspectives Guide on AI Safety. • This guide is the first step toward realizing safe, secure, and reliable AI. Guide to Evaluation Perspectives on AI Safety https://aisi.go.jp/assets/pdf/ai_safety_eval_v1.01_en.pdf AISIJapan AI Safety Institute
11. SC 42 – Artificial Intelligence Guide to Evaluation Perspectives on AI Safety Key Elements of AI Safety Human-Centric Safety Fairness Ensuring Security Privacy Protection Transparency Perspectives for AI Safety Evaluations AI Guidelines for Business and survey results on international publications and tools. Control of Toxic Output Fairness and Inclusion for Users Privacy Protection Explainability Data Quality Prevention of Misinformation, Disinformation and Manipulation Addressing to High-risk Use and Unintended Use Security Robustness Verifiability AISIJapan AI Safety Institute
12. • The significance and use of the Red Teaming Methodology Guide on AI Safety. Red Teaming Methodology Guide https://aisi.go.jp/assets/pdf/ai_safety_RT_v1.00_en.pdf AISIJapan AI Safety Institute
13. SC 42 – Artificial Intelligence Activity Map on AI Safety –Summary- Safety Inclusivity Innovation Automated Evaluation External Testing Threat Actor Uplift Evaluation Watermarking Hallucination Alignment Content Provenance Disclaimer AI Label Sustainability AGI AI-agent GPAI Public Sector Manufacturing, Robotics and Mobility Logistics and Healthcare Government SMEs and Startups Privacy AI for Critical Infrastructure IT/OT Management System Ethics Trustworthiness Accountability Fairness 人間中心 Ensuring Fair Competition Cognitive and Behavioral Manipulation Diversity Robustness Multi-Stakeholder Social Resilience Investments in Education Investments in Individuals AI Utilization Academic Research Conformity Assessment Regulations and Law Responsible AI Development セーフガー ド/ガード レール Disinformation Interdiscipli nary Collaborati on Interoperability Incident Response and Sharing among Industry, Academia and Government Multicultural Mutual Understanding AI safety washing OP: Originator Profile Grants and Startups by Government Vertical Standard Disinformation and Content Authentication International Cooperation Human Capital Investment Research Investment Ensuring Transparency Accessibility and Safety Net Social Challenges on Sustainability AI Safety Evaluation and Red-teaming Governance Framework Risk-based Approach Risk Mitigation Dual-use foundation model Transparency Report Model Card System Card Data Card CBRN Foundation Model Domain- specific Foundation Model Safety for Emerging Technology Dual-use Output Attribution Enhancing Interpretability Accreditation and Certification System Biological and Chemical Sharing Vulnerabilities and Organizational Response Regulations for Eco System Domain Investment Version: v10 Update: 24-08-23 Content Authentication Physical, Cyber and Domestic Threats Protecting Personal Information and Intellectual Properties High-risk AI System Data Quality Traceability Action Sub-action Profiling
14. SC 42 – Artificial Intelligence Related activities
15. • Identifying Areas of Strength in Japan (Example.) • Strengthen information dissemination • Partnerships with independent and national research institutes, etc. • Collection and publication of best practices and related case studies. • Organize cooperation with private companies Future Planned Initiatives
16. SC 42 – Artificial Intelligence Summary AISIJapan AI Safety Institute
17. SC 42 – Artificial Intelligence Thank you AISIJapan AI Safety Institute