There was an interesting piece in the Washington Post on Tuesday discussing the growing use of HireVue. HireVue is used to screen applications for jobs. It appears to do this both by administering tests and "games" to applicants (which was common pre-AI), and by holding a video screening event with candidates, where an algorithm analyzes the candidate's behavior, and determines whether they are suitable for the job.
This system raises many ethical questions, and the Post article goes over these in some detail. However, it seems to me that it has even more extreme legal questions. For example, in the United States, it is illegal to use a hiring test that does not directly pertain to the applicant's ability to perform the job.
I am curious about whether the construction of an AI system that can meet this kind of legal standard is possible, and what such a system would have to look like, especially given the history of AI systems merely learning to copy the discriminatory patterns in our society when applied to problems like this one.
Are legal AI systems for hiring likely to be possible? If so, what are they likely to look like?