The U.S. Visa May Soon Scan New Immigrants’ Faces, Irises, Voices, and DNA

If enacted, the personal information of more than 70% of those applying for immigration will be entered into a DHS database.

5 min readDec 20, 2020

OneZero’s General Intelligence is a roundup of the most important artificial intelligence and facial recognition news of the week.

The Department of Homeland Security is looking to scan the faces, irises, voices, and DNA of millions more people per year, according to new rules proposed by the agency.

DHS claims that the collection of data from children, especially their DNA, is meant to help fight human trafficking at the border by verifying that children are related to the adults transporting them across U.S. borders.

The DHS rules explicitly say that biometric data obtained through the immigration process would be kept and shared with law enforcement.

These new rules put immigrants under a form of algorithmic surveillance. Being in a law enforcement database puts people not suspected of any wrongdoing as a possible endpoint for an algorithm looking for a match in a criminal investigation. One poor-quality image uploaded for a facial recognition match could put a new immigrant under suspicion based on a faulty search.

Algorithms like the ones used by DHS and federal law enforcement are also subject to bias on racial and gender lines, meaning they are more likely to serve incorrect matches when the subject is not white and male.

These biometric tests don’t just expand the surveillance of immigrants — they are also incredibly costly. DHS says that this new initiative could cost up to $500 million per year, which includes nearly $160 million in new testing costs, as well as a new system to submit DNA to DHS for testing and storage.

These rules are open to comments for 60 days, after which DHS will review them and take action.

And now, here is some of the most interesting A.I. research of the week.

Testing GPT-3’s proficiency at law (and 56 other things)

Language algorithms are slowly learning to multitask. For instance, the same algorithm that facilitates auto-complete on a virtual keyboard can be used to answer questions based on a paragraph, or to translate text into another language. To test how good algorithms are at doing not just one task, but many, researchers created a set of 57 tasks for multipurpose language algorithms. Most models aren’t very good, but there’s one clear standout: OpenAI’s new GPT-3 algorithm scored higher than the competition, at times 20% higher than random chance.

Why isn’t there better A.I. in video games?

Video game characters might seem artificially intelligent: They can walk around, accomplish things in the digital world, and respond to players’ actions. But usually, they’re just following a script. Microsoft interviewed 17 game developers who specialize in A.I. around the industry to figure out why this is still the case. One challenge turned out to be that it’s hard to make an A.I. algorithm that’s fun to play against at different difficulty settings. One developer said it was difficult to train the algorithm to shoot poorly on purpose in order to cater to less skilled players, saying it would confuse the training process.

Critique a la critique

Michael J. Lyons, co-author of a dataset called JAFFE that’s currently used for facial recognition purposes, wrote a confrontational response to the inclusion of his dataset in an art show called Excavating A.I. put on by Kate Crawford and Trevor Paglen. Lyons accuses Crawford and Paglen of using the images for commercial, non-research purposes — and of exploiting the images of people that appear in the dataset the same way facial recognition algorithms do.

The U.S. Visa May Soon Scan New Immigrants’ Faces, Irises, Voices, and DNA

If enacted, the personal information of more than 70% of those applying for immigration will be entered into a DHS database.

Testing GPT-3’s proficiency at law (and 56 other things)

Why isn’t there better A.I. in video games?

Critique a la critique

Written by Alizabeth2riya

No responses yet