Ito and I saw the play “Data” at Arena Stage last Tuesday. The Washington Post review was titled “This play is a flawed look at AI. You should see it anyway.” “Data-Arena Stage review” We agree. The production and acting were outstanding. But the story fell short in a number of ways.
Skipping all the personal mysteries that were inadequately explained, the play suggests that turning the screening of immigration applications (not asylum applications) over to a computer (AI) program would be bad. Whatever biases (criteria) are wanted for US immigration policy can be built into the screening program, of course, but if they are not the criteria America wants to apply, AI is a safer way of avoiding them than the judgement of individual immigration officers.
In requesting bids from programmers to develop the AI screening program, the play states that the government’s objective to sort out those applicants for residency (and ultimately citizenship) is to approve those who would be “positive and productive.” If the criteria for finding such people can be identified for immigration officers (no easy task), they can be built into an AI program, which can be relied on to more faithfully and consistently apply them than any human officers. The boss in the play correctly noted that such a program could produce an answer in seconds that took the US immigration service three years to achieve in his case coming from China.
Of course, AI is not perfect and can make mistakes just as humans can. But their accuracy is improving with training and use at a rapid rate. Tesla’s Full Self Driving cars have been linked to 956 accidents with 29 deaths. But this is already dramatically safer than the much higher death rates per million miles driven of car accidents by humans.
Training AI programs will draw on a much wider set of information than is now used for immigration applications. Anything on the Internet related to the applicant might be collected and evaluated by an AI program. How should such information be used? This opens new concerns that will need to be evaluated, but the promise of faster and better application processing from the use of AI promises much more benefits than risks compared to our current reliance on human immigration officers.
“Data” does a poor job of exploring these important issues, but it is worth watching anyway.
Warren's space 
I think it would benefit USA to raise the productivity of its exising stock of immigrants as well as new immigrants by setting a qualification exam that assessed the merits of the immigrant / refugee. An immigrant would be granted entry if they achieved a score higher than the current President of the zUnited States. The applicant could be given two years to write the exam which would grade proficiency in English and appreciation of American cultural values. If the immigrant failed the exam they would be relocated to their home country and not allowed to reapply for two years.