From polygraphs to modern lie detectors: what past failures can teach us today
From the 15th edition of my newsletter, Humane AI
Summer with all its traveling is over! This issue of my newsletter, Humane AI is about traveling, inspired by my summer vacation. But, more importantly, writing this issue is inspired by a passport-control officer in an airport in Varna, Bulgaria, skeptically scratching my Schengen visa with the edge of a coin, to make sure that the visa sitting confidently in my Iranian passport is actually real!
For the next few minutes, I invite you to read about AI being used in immigration and traveling, geo-tagging and its negative impacts on the environment, Airbnb’s “Experiences” and more.
✏ From polygraphs to AI-enabled lie detectors: they all lie but are still in use. This time, for travelers!
“Take a quarter out of your pocket. Flip it, if it’s heads, you are lying; if it’s tails, you are telling the truth… That’s exactly the same degree of accuracy that this polygraph instrument has.” Said by Doug Williams, a former Oklahoma City police detective who ran polygraph examinations between 1972 to 1979. Later, he decided to spend his life training people to pass the test via different techniques including calming themselves by picturing being on the beach, thinking about their greatest fears, or even clenching their anus!
Despite the scientific community’s concerns about the lack of accuracy of these tools, law enforcement agencies still use polygraphs for finding criminals, spies, etc. According to this Intercept article, an AI-enabled lie-detector — a virtual policeman piloted at the Hungary, Greece and Latvia airports — is there to ask you questions at the border. “You provide your answers verbally, and the virtual policeman uses your webcam to scan your face and eye movements for signs of lying.” The project was developed by researchers at Manchester Metropolitan University, “who say that the technology can pick up on ‘micro gestures’ a person makes while answering questions on their computer, analyzing their facial expressions, gaze, and posture.” The article raises concerns about the accuracy and transparency of this tool. To learn more about the study involved, read this paper. The number of people who participated in the study to train and test this tool is 32; the overall accuracy, 75%; and the funding supplied by the EU, 4.5 million euros!
📎 If you want to dig into studies about using machine learning algorithms to infer one’s well-being, general intelligence, writing ability, verbal fluency, sexual orientation, political orientation, personality traits, autism, depression, dyslexia, psychopathy, and stress, check out this paper entitled “Can Machines Read our Minds? “ published by the Oxford Internet Institute. Researchers examined 26 studies on the possibility of predicting psychological traits and mental states of individuals via their social media activities, facial features, etc. The paper also raises questions about the issue of trust, informed consent, and digital identity.
✏How the U.S. Department of Homeland Security uses digital data in the name of national security
The idea of monitoring social media profiles to grant an entry visa to the U.S. is not a new thing (see this timeline). However, last week, social media monitoring took an Orwellian turn by learning that a Palestinian freshman Harvard student was deported because of his friend’s facebook post. Read the story here. The authors of the paper, Social Media Monitoring, questioned the effectiveness of collecting social media data for the sake of border security. Also, according to the report, “DHS’s social media monitoring pilot programs seem to have focused in large part on Muslims: at least two targeted Syrian refugees, one targeted both Syrian and Iraqi refugees, and the analytical tool used in at least two pilots was tailored to Arabic speakers.” According to the report, one of the impacts of such initiatives has been “self-censorship of people applying for visas as well as their family members and friends.”
📎 For those of you interested in law, I suggest reading this great Lawfare article entitled “The National Vetting Enterprise: Artificial Intelligence and Immigration Enforcement” by Chinmayi Sharma. The article goes into the details of the Department for Homeland Security proposals for using AI in immigration. It breaks down the specific roles that USCIS, ICE, and CBP play in executing DHS orders.
Why is this so important to me? I’ll tell you: because of the nature of my work (relating to human rights implications of new technologies) I have decided not to travel back to my country in order to avoid any kind of potential governmental hostility toward dual citizens. For fourteen days out of all these 365 days in a year, my mother, my sister and I try to re-unite somewhere outside Iran. We plan months before, get excited, go through visa checklists and interviews over video calls. But then I see them getting rejected again and again, barring them from entering the U.S. (due to the Muslim Ban) and all European countries in the Schengen zone. Sometimes, I wonder what is the value of all these “reliable vetting systems” and “safe border technologies” if, in the end, their collective decision is to ban this: the joy of being with your family and being safe.
✏How geotagging on Instagram has been ruining environment — and taking human lives?!
“When we search for London on social media, it’s the first thing we see,” said a South Korean visitor to the Seven Sisters cliffs in England. Because of social media geo-tagging, influencer posts, and search algorithms, the Seven Sister cliff — formerly known only to avid hikers — has become one of the favorite spots for tourism. According to the article “the growing number of visitors has raised concerns about the risks of having so many people high up on the cliffs. Last year, Kim Hye-won, a Korean student, died after losing her footing there.” According to this Change.org petition, “when photos of a location gain popularity on Instagram (via the geotag and hashtag features, plus shares or ‘features’), the result is often a surge in visitation. Unfortunately, a number of outdoor destinations all over the world have been closed or destroyed due to the negative impacts from this influx of visitors.”
Signatories of the Change.org petition ask Facebook/Instagram to allow users to “report violations that are both illegal and harmful to the environment” and implement a better content moderation practices to curb the virality of photos/accounts which promote harmful practices.
✏Machine Learning-powered search ranking of Airbnb Experiences
This summer, during my trip, I noticed that Airbnb has started a service called Experiences to help locals teach classes to guests or take them on local tours. This post published on the Airbnb Engineering and Data Science Medium page goes into details about Airbnb’s plan to better personalize this service. Here is an excerpt: “We can use the country information to personalize the Experience ranking based on Categories preferred by users from those countries. For example, historical data tells us that when visiting Paris Japanese travelers prefer Classes & Workshops (e.g. Perfume making), US travelers prefer Food & Drink Experiences, while French travelers prefer History & Volunteering. We used this information to engineer several personalization features at the Origin — Destination level.”
Okay, I’ll skip over expressing my speculation about this service’s potential discriminatory impacts because of training on biased historical data, or the effects on local businesses, or potential concerns about treating different demographics in different ways — note that Airbnb has already been struggling to address racial discriminations in its services. I won’t even go there; I just want to highlight this sentence from the Medium post: “US travelers prefer Food & Drink Experiences, while French travelers prefer History & Volunteering?!! This made me burst out laughing! What can I say my American friends? ¯\_(ツ)_/¯ I just know that this sentence, all by itself, has the potential to trigger a new “Freedom Fries” proposal!
This has been an excerpt from the Humane AI newsletter #15, run by me, Roya Pakzad. You can subscribe to the newsletter here.
