算法型评分工具:优势、风险与法律规制

Algorithm ScoringTool:Advantages,Risks,and Legal Regulations

在个人金融借款业务中,对借款人进行信用评价是风险控制的核心手段。信用评价的工具随着科技进步而不断改进,近年来又兴起了向算法型评分工具发展的趋势。算法型评分工具的特点表现为,依托机器学习算法和以借款人的新型替代数据为分析对象。之所以要引入算法型评分工具,是因为其相较于传统型评分工具具有增强包容性和提高预测准确性的显著优势,符合信用评分行业准确性和公平性的核心要求。与此同时,伴随着机器学习算法的嵌入,算法型评分工具自身也潜藏着评分不透明、不准确和不公平的风险。在我国现行法律体系下,《个人信息保护法》提供了可用于规制上述风险的法律手段。具体来说,信用评分机构须主动就其算法开展贯穿全周期的影响评估,评估活动应遵循较为严格的评估标准;借款人可向信贷机构或评分机构主张行使算法解释权,评分机构的商业秘密保护和算法的技术复杂性并不会影响该权利的可行性和实效性;借款人还可以信贷机构为对象主张算法结果拒绝权,要求信贷机构对贷款决策进行人工干预。

In personal financial loan business, credit evaluation of borrowers is the core means of risk control.The tools for credit evaluation have been continuously improved with the advancement of science and technology,and in recent years, there has been a trend towards algorithm-based scoring tools. Algorithmic scoring tools are characterized by relying on machine learning algorithms and using new alternative data of borrowers as the object of analysis. The reason for introducing algorithmic scoring tools is that compared with traditional scoring tools, it has significant advantages in enhancing inclusiveness and improving prediction accuracy, and meets the core requirements of accuracy and fairness in the credit scoring industry. At the same time, with the embedding of machine learning algorithms, the algorithmic scoring tools themselves also have the potential to be opaque, inaccurate and unfair in scoring. Under China’s current legal system, the "Personal Information Protection Law" provides legal means that can be used to regulate the above risks. Specifically, credit scoring agencies must take the initiative to carry out full-cycle impact assessments on their algorithms, and assessment activities should follow stricter assessment standards;Borrowers can claim to the credit institution or the scoring agency to exercise the algorithm interpretation right, and the commercial secret protection of the scoring agency and the technical complexity of the algorithm will not affect the feasibility and effectiveness of the right;Borrowers can also claim the right of refusal of algorithm results for credit institutions, requiring credit institutions to manually intervene in loan decisions.