Not Deletion,Only Forgetfulness:A Purposeful Interpretation to the Deletion of Personal Information in AI Large Models
The obligation to delete personal information is pivotal in safeguarding data subjects'autonomy in our increasingly digital world.The proliferation of AI technology and big data has significantly expanded personal information collection,intensifying privacy concerns.This paper explores the substantial challenges and complexities faced by large AI models,in complying with personal information deletion obligations,addressing both the technical hurdles and the evolving legal landscape.From a technical perspective,large language models(LLMs)like ChatGPT encounter diverse challenges in the deletion of personal information.These include the unknowability of data embedded within complex neural networks,the uncontrollability of sophisticated,evolving algorithms,the unverifiability of complete data erasure,and the prohibitively high costs associated with comprehensive data deletion processes.These challenges are exacerbated by the intrinsic characteristics of Al systems,which are designed to continuously learn and adapt,making it increasingly difficult to ensure that all personal information is irrevocably removed from these complex systems.On the legal front,the intricacies of the Personal Information Deletion Clause in Article 47(2)of the Personal Information Protection Law in China present a significant challenge.The clause's lack of alignment with the operational principles of AI models results in vague and ambiguous guidelines for executing deletion obligations,creating compliance difficulties.Additionally,the EU's recent Artificial Intelligence Act fails to address this issue directly,necessitating careful interpretation in conjunction with Article 17 of the General Data Protection Regulation(GDPR)and other relevant regulations.The Regulation on the Management of Generative AI Services(RMGAI),effective from August 15,2023,further add to this complexity with varying terminology,potentially leading to inconsistent interpretations and enforcement challenges in the AI sector.Internationally,the standards for AI models'obligation to delete personal information are diverse and often conflicting.The United States Federal Trade Commission's focus on algorithmic deletion as an essential tool for AI regulation has sparked intense debate and could result in restrictive policies and developmental constraints for AI industries,especially in less developed nations.This global disparity underscores the need for a more balanced,universally applicable approach to AI governance.This paper uniquely combines legal and information science perspectives to analyze the technical features of AI models in processing personal information.It scrutinizes the full-process risk monitoring and the complex interplay between compliance and liability mitigation,offering a critical perspective against the dominant American approach.The paper advocates for a comprehensive,multidimensional solution that harmonizes technical feasibility with legal and ethical considerations.The paper significantly contributes to the academic discussion in four areas.First,it provides a detailed and systematic examination of AI models'obligations to delete personal information,blending legal interpretative techniques with technical insights.Second,it fills an existing gap in literature concerning AI models'passive obligations in personal information handling.This novel approach integrates AI operational principles with data protection imperatives,offering a distinctive perspective.Third,the paper acknowledges the inherent limitations of AI models in fulfilling deletion obligations.It proposes a purpose-driven interpretation of these obligations under the Personal Information Protection Law,diverging from the conventional debates on deletion rights and the right to be forgotten,and emphasizes more pragmatic,feasible solutions.Fourth,the paper delves into the RMGAI regarding AI models'personal information deletion processes.It suggests a comprehensive approach,including establishing alternative deletion methods and exploring liability reduction for cases where deleted personal information reappears,thereby contributing to a more nuanced understanding and application of these regulations.
AI large modelspersonal informationobligation to deletepurposeful interpretationanonymizationresponsibility gaps
万方数据
