Human Oversight And Risk-Based Approach To AI

Human Oversight And Risk-Based Approach To AI Presented by: Dr. Kowalczewska Date: 13 June 2024 Introduction - Overview of AI regulation as a global concern. - Common objectives: Innovation and safeguarding humanity. - Focus on EU's AI Act and LAWS discussions within the Convention on Certain Conventional Weapons (CCW). AI Act: Ethical Foundations - Emphasis on human rights, democracy, and rule of law. - Ethical principles from AI High-Level Expert Group (AI HLEG). • Human agency and oversight. • Technical robustness and safety. • Privacy and data governance. • Transparency. • Diversity, non-discrimination, and fairness. • Societal and environmental well-being. AI Act: Risk-Based Approach - Categorization of AI systems by risk: unacceptable, high, limited, minimal. - Prohibition of unacceptable AI uses (e.g., manipulative techniques, social scoring). - High-risk AI systems: rigorous requirements (e.g., human oversight, transparency) AI Act: Exclusions - Exclusion of military, defense, or national security applications of AI (by public or private entities) • However, if subsequently used outside of its military purpose, temporarily or permanently, such as for civilian, humanitarian, or law enforcement purposes, it falls back within the scope of the AIA. • Procedural issue (art. 4(2) TEU and Chapter 2 of Title V of the TEU) • Normative framework remains similar - ethical underpinnings + fundamental rights and PIL AI Act: General Purpose Regulation - Solid starting point to be complemented by targeted regulations - Main burden on AI providers (procedural requirements) - Use-case language (not technology-type language) - Basis for the regulation of trustworthy AI LAWS: AI-Framed Definition - LAWS refer to AI systems intended for military purposes, which exhibit adaptability post-deployment by operators. These systems are capable of inferring, from received inputs, how to generate outputs such as decisions regarding the identification and selection of military targets, which may influence the physical environment through engagement with military targets (including people or objects), potentially resulting in serious incidents. LAWS: Is Current Law Enough? - IHL compliance: Distinction, proportionality, precautions. - Ethical considerations: Accountability gap, human control over lethal decisions. - My argument: states need to pronounce their stance on the acceptable level of risk to fundamental rights, especially the right to life, within the context of armed conflict. • Adopt a risk-based approach, akin to AIA, by explicitly prohibiting certain uses of LAWS and regulating high-risk LAWS more precisely with a set of mitigating measures LAWS: Supporting propositions - Two-tier approach (July 2022) • - Finland, France, Germany, the Netherlands, Norway, Spain, and Sweden. "Draft Articles on Autonomous Weapons" (May 2023) • Australia, Canada, Japan, the Republic of Korea, Poland, the United Kingdom, the United States - Unacceptable risk • Operate entirely outside of human control and a responsible chain of command • Designed to contravene IHL principles Convergence in AI Regulation - Common ethical principles in AIA and LAWS: Human dignity, oversight, risk management. - AIA serves as a directive targeting economic entities, mandating compliance for profit generation within the EU, its adoption is relatively straightforward compared to the negotiation and implementation of a multilateral arms treaty. • If EU MS are committed to upholding the rule-based order, they should actively articulate, in a legally binding manner, the unacceptable risks posed by AI in armed conflicts