关键词：
Computer science
Ethics
Society
摘要：
[...]a modeller wishing to explore whether ethnicity would affect treatment recommendations made online by health-care professionals could create two patient profiles that differ only in some respect that relates to ethnicity.
Integration of expert knowledge, particularly from the social sciences and including qualitative methods, can help to overcome such limitations.
A causal model allows us to ask and answer questions such as 'Had the past been different, would the present or future have changed?' In the visa example (see 'Three causal tests', part 1), algorithmic biases could be smoked out by tweaking parts of the model to explore, for instance: 'Had individual X been Christian, would this algorithm have granted them a visa?' A researcher could then identify what pieces of information an algorithm could use to achieve counterfactual fairness4: the algorithm's output would not change regardless of the individual's religion.
[...]the models can help regulatory agencies to understand how changing a scholarship algorithm influences who is accepted into law school (see 'Three causal tests', part 3).
欢迎来到三峡大学图书馆！
