(English Version: Top-right translate button OR scroll down.)

最终判断和理解责任未被消除

AI通过模拟预测交易和持续追踪交易可以提高交易的可解释性，帮助用户理解交易。但是，AI不能替用户、开发者和平台团队完成带责任的最终判断。

对低频、新型和复杂异常的处理难题未被消除

通过AI的动态监测和自动化策略可以处理一些高频、重复的异常，这些普遍的异常被模型化后便能启动自动化程序去响应。但是针对缺乏历史样本的新型攻击或异常，针对难以靠固定规则识别的跨上下文、跨协议的复杂异常，AI识别能力有限，即使识别到，也不一定能定义或自动决定最优响应。

对数据质量和规则定义的依赖未被消除

AI的判断标准通常建立在模型、输入数据、情报来源和预设规则上。但是AI的判断质量仍然依赖上游数据质量、情报来源和规则定义，它不能自动保证输入本身真实、完整、无偏。

The Responsibility for Final Judgment and Understanding Has Not Been Eliminated

AI can improve the interpretability of transactions and help users understand them through simulated transaction prediction and continuous transaction tracking. However, AI cannot complete the final judgment involving responsibility on behalf of users, developers, and platform teams.

The Difficulty of Handling Low-Frequency, Novel, and Complex Anomalies Has Not Been Eliminated

AI's dynamic monitoring and automated strategies can handle some high-frequency, repetitive anomalies; once these common anomalies are modeled, automated programs can be launched to respond. However, for novel attacks or anomalies that lack historical samples, and for complex cross-context, cross-protocol anomalies that are difficult to identify by fixed rules, AI's identification capabilities are limited. Even if identified, AI cannot necessarily define or automatically decide on the optimal response.

The Dependence on Data Quality and Rule Definitions Has Not Been Eliminated

AI's judgment standards are typically built on models, input data, intelligence sources, and preset rules. However, the quality of AI's judgment still depends on the quality of upstream data, intelligence sources, and rule definitions. It cannot automatically guarantee that the input itself is truthful, complete, and unbiased.