Abstract:

Big data algorithms are not valueless. Social goals are inevitably embedded in the process of algorithm development.And it brings new problems such as algorithm bias and even algorithm discrimination at the same time. Algorithm bias is a very hidden intelligent copy of social bias, including the bias in training data, the bias of algorithm designers, and the bias of the algorithm itself. It is necessary to further distinguish algorithm bias and algorithm discrimination, prevent political embedding and capital intervention in algorithm design, and avoid big data algorithms becoming tools for interest groups and power departments to achieve economic, social and even political goals. At the same time, effective algorithmic supervision and review mechanisms should be employed to ensure and promote future social justice through algorithmic justice.

Key words: big data, algorithm bias, algorithmic discrimination, value load

摘要：

大数据算法并不是价值无涉的,算法开发过程中不可避免地嵌入了社会性目标,带来了所谓的算法偏见甚至算法歧视等新问题。算法偏见是社会偏见非常隐蔽的智能复制,包括训练数据中的偏见、算法设计者的偏见,以及算法本身的偏见等不同形式。必须进一步区分算法偏见和算法歧视,防止算法设计中的政治内嵌和资本介入,避免大数据算法沦为利益集团和权力部门实现经济、社会甚至政治目标的工具,同时辅之以有效的算法监督和审查机制,以算法正义保证并推动未来的社会正义。

关键词: 大数据, 算法偏见, 算法歧视, 价值负载