人脸情绪识别算法的风险与规制

The Risk and Regulation of Facial Emotion Recognition Algorithm

人脸情绪识别算法是人脸识别的特殊变体,也是情感计算的重要构成。由于技术的更迭、应用的泛滥和风险的特殊,这一算法在实践中存在大量隐患,亟待法律规制。在美国、欧盟和我国的法律框架下,人脸情绪识别算法目前可以通过"算法"和"数据"两条路径进行部分规制。然而,算法本身的特殊性和立法技术的局限性,决定了还有大量风险无法得到回应,更难以实现权利保障和技术创新的平衡。为此,应当引入"风险为本"的规制思路:以人脸情绪识别算法在不同场景下的正当性、安全性、准确性和可责性四类风险作为法律规制的核心;通过规则设计,实现算法部署的价值公允、算法运行的数据安全、算法决策的准确可靠和算法追责的透明便捷。

The facial emotion recognition algorithm is a particular variant of facial recognition and is an essential component of affective computing. Due to the change of technology, the proliferation of applications and the unique risks, the algorithm is triggering significant concern in practice, which calls for regulation urgently. Based on the current framework of the United States, the European Union, and China, the algorithm can be partially regulated via the path of algorithms or data. However, the particularity of the algorithm and the drawbacks in legal technique determine that there are still many risks. And the balance between governance and innovation could hardly be achieved. To this end, the “risk-based” regulation should be introduced. The four types of risk of the legitimacy in deployment, safety in operation, accuracy in decision-making, and accountability in remedy are taken as the core of the regulation in different scenarios. Thus the fair value of algorithm deployment, the data security of algorithm operation, the accuracy and reliability of algorithm decision-making, and the transparency and convenience of algorithm accountability could be realized through the regulation design.