Can there be responsible AI without AI liability? Incentivizing generative AI safety through ex-post tort liability under the EU AI liability directive
{"title":"Can there be responsible AI without AI liability? Incentivizing generative AI safety through ex-post tort liability under the EU AI liability directive","authors":"Guido Noto La Diega, Leonardo C T Bezerra","doi":"10.1093/ijlit/eaae021","DOIUrl":null,"url":null,"abstract":"In Europe, the governance discourse surrounding artificial intelligence (AI) has been predominantly centred on the AI Act, with a proliferation of books, certification courses, and discussions emerging even before its adoption. This narrow focus has overshadowed other crucial regulatory interventions that promise to fundamentally shape AI. This article highlights the proposed EU AI liability directive (AILD), the first attempt to harmonize general tort law in response to AI-related threats, addressing critical issues such as evidence discovery and causal links. As AI risks proliferate, this article argues for the necessity of a responsive system to adequately address AI harms as they arise. AI safety and responsible AI, central themes in current regulatory discussions, must be prioritized, with ex-post liability in tort playing a crucial role in achieving these objectives. This is particularly pertinent as AI systems become more autonomous and unpredictable, rendering the ex-ante risk assessments mandated by the AI Act insufficient. The AILD’s focus on fault and its limited scope is also inadequate. The proposed easing of the burden of proof for victims of AI, through enhanced discovery rules and presumptions of causal links, is insufficient in a context where Large Language Models exhibit unpredictable behaviours and humans increasingly rely on autonomous agents for complex tasks. Moreover, the AILD’s reliance on the concept of risk, inherited from the AI Act, is misplaced, as tort liability intervenes after the risk has materialized. However, the inherent risks in AI systems could justify EU harmonization of AI torts in the direction of strict liability. Bridging the liability gap will enhance AI safety and responsibility, better protect individuals from AI harms, and ensure that tort law remains a vital regulatory tool.","PeriodicalId":44278,"journal":{"name":"International Journal of Law and Information Technology","volume":"106 1","pages":""},"PeriodicalIF":1.6000,"publicationDate":"2024-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"International Journal of Law and Information Technology","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1093/ijlit/eaae021","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"LAW","Score":null,"Total":0}
引用次数: 0
Abstract
In Europe, the governance discourse surrounding artificial intelligence (AI) has been predominantly centred on the AI Act, with a proliferation of books, certification courses, and discussions emerging even before its adoption. This narrow focus has overshadowed other crucial regulatory interventions that promise to fundamentally shape AI. This article highlights the proposed EU AI liability directive (AILD), the first attempt to harmonize general tort law in response to AI-related threats, addressing critical issues such as evidence discovery and causal links. As AI risks proliferate, this article argues for the necessity of a responsive system to adequately address AI harms as they arise. AI safety and responsible AI, central themes in current regulatory discussions, must be prioritized, with ex-post liability in tort playing a crucial role in achieving these objectives. This is particularly pertinent as AI systems become more autonomous and unpredictable, rendering the ex-ante risk assessments mandated by the AI Act insufficient. The AILD’s focus on fault and its limited scope is also inadequate. The proposed easing of the burden of proof for victims of AI, through enhanced discovery rules and presumptions of causal links, is insufficient in a context where Large Language Models exhibit unpredictable behaviours and humans increasingly rely on autonomous agents for complex tasks. Moreover, the AILD’s reliance on the concept of risk, inherited from the AI Act, is misplaced, as tort liability intervenes after the risk has materialized. However, the inherent risks in AI systems could justify EU harmonization of AI torts in the direction of strict liability. Bridging the liability gap will enhance AI safety and responsibility, better protect individuals from AI harms, and ensure that tort law remains a vital regulatory tool.
在欧洲,围绕人工智能(AI)的治理讨论主要集中在《人工智能法案》上,甚至在该法案通过之前就出现了大量的书籍、认证课程和讨论。这种狭隘的关注掩盖了有望从根本上塑造人工智能的其他重要监管干预措施。本文重点介绍了拟议中的欧盟人工智能责任指令(AILD),这是为应对人工智能相关威胁而协调一般侵权法的首次尝试,解决了证据发现和因果联系等关键问题。随着人工智能风险的激增,本文认为有必要建立一个反应迅速的系统,以充分应对人工智能损害的出现。人工智能安全和负责任的人工智能是当前监管讨论的核心主题,必须优先考虑,而事后侵权责任在实现这些目标方面发挥着至关重要的作用。随着人工智能系统变得更加自主和不可预测,《人工智能法》规定的事前风险评估就显得不够充分,因此这一点尤为重要。人工智能法》对过错的关注及其有限的范围也是不够的。在大型语言模型(Large Language Models)表现出不可预测的行为以及人类越来越依赖自主代理执行复杂任务的情况下,通过加强发现规则和因果联系推定来减轻人工智能受害者举证责任的建议是不够的。此外,《人工智能法》对风险概念的依赖是错误的,因为侵权责任是在风险发生后才介入的。然而,人工智能系统固有的风险可以证明欧盟协调人工智能侵权行为的严格责任方向是正确的。弥合责任差距将加强人工智能的安全和责任,更好地保护个人免受人工智能的伤害,并确保侵权法仍然是一个重要的监管工具。
期刊介绍:
The International Journal of Law and Information Technology provides cutting-edge and comprehensive analysis of Information Technology, Communications and Cyberspace law as well as the issues arising from applying Information and Communications Technologies (ICT) to legal practice. International in scope, this journal has become essential for legal and computing professionals and legal scholars of the law related to IT.