瞭望 | AI“向真”须严防数据“投毒”

Core Insights - The emergence of Generative Engine Optimization (GEO) is leading to data poisoning behaviors that compromise the integrity of AI-generated information [1][2] - Data poisoning can undermine information fairness, harm user rights, and hinder healthy industry development [1][2] Group 1: Data Poisoning Risks - Data poisoning disrupts information fairness by amplifying false information, causing quality content to be overshadowed [1] - Users may make erroneous decisions based on non-objective information, particularly in high-stakes areas like finance and healthcare, potentially leading to financial loss or safety risks [1] - The repeated citation of incorrect information in AI models can erode user trust in AI, negatively impacting innovation and development quality in the AI industry [1] Group 2: Mitigation Strategies - Government departments should enhance regulatory guidance and establish industry standards related to GEO, focusing on data source verification, quality assessment, and content authenticity [2] - Companies must strengthen technical self-discipline, improve data screening processes, and develop high-precision techniques for identifying and filtering toxic data [2] - Public awareness of AI technology and the ability to discern false information should be improved, encouraging feedback on AI anomalies to foster a healthy AI governance ecosystem [2]