论医疗人工智能“黑箱”难题的应对策略和规制进路

On the countermeasures and regulatory approaches to the black box of medical artificial intelligence

人工智能(AI)在医疗领域具有重大应用价值,不过,“黑箱”属性是其走向广泛应用的最大阻碍。通过法律制度的合理设计可以有效应对医疗AI的“黑箱”难题。转变规制思路,重点运用临床试验、真实世界研究等其他方法管控医疗AI之风险。从对算法解释的规制转向对算法透明和可追溯的规制,从对算法本身的规制转向对算法之外的其他环节和要素的规制,并契合医疗AI“非锁定”之特性,可建立全生命周期的规制体系。同时,平衡好监管安全和技术创新之关系,防止规制过度,应适度“温和”上市前的监管措施,在特定情形下豁免审批或加速审批,将监管重心适当后移,为医疗AI之发展创造合理空间。

Despite its great medical value,the black-box of AI is the biggest concern that prevents it from wide application. Scientific understanding is not the goal of drug and equipment supervision,but a replaceable means to achieve the goal itself,which means we could deal with the issue of black-box by law. We can control the risk of AI through alternative the ways of utilizing clinical trial and real-world research. We can shift the focus from the regulation of algorithm interpretation to that of transparency and traceability,from the regulation of algorithm to segments and elements beyond algorithm,and accordingly establish a system of total product life cycle. In the meantime,to encourage AI development and innovation,we can simplify the approval procedures in some cases and focus on the market after rather than before its availability.