Rafael Conejos (George Washington U Law) has posted “Social Media Should be a Scrape Free Zone from AIFS” on SSRN. Here is the abstract:
Artificial Intelligence (AI) is used to screen job applicants in the United States (U.S.) everyday by the thousands yet neither the applicants nor some of the employers who use it know precisely how they work or why they arrive at those results. This is due to algorithms being trade secrets and the black box dilemma of AI. If rejection is based on factors which amount to unlawful discrimination, how does an applicant prove it if the process is a secret?
In a highly competitive job environment, where there are more applicants than there are openings, employers rely on the predictions made by Artificial Intelligence Filtering System (AIFS) in determining, based on limited data, if one applicant will underperform or outperform another applicant. AIFS achieves this through training data which instructs it to seek out desirable qualities in an applicant from the information the latter has provided or which AIFS finds publicly available. The more information it has about an applicant, the more accurate the prediction. Like a general warrant being served by an officer in a home, AIFS scrape all publicly available information about the applicant, including their social media accounts, with the goal that more information, regardless of relevance for the role, is helpful when it ‘scores’ one applicant against another.
However, not all social media accounts are intended by the applicant to be viewed through the lens of an employer’s AIFS. Often, social media, outside of LinkedIn, is a place of self-expression and public advocacy. There are many private matters in one’s life which an employer has no business in using to assess one’s fitness for a role. Allowing AIFS to remove the practical obscurity of a person by scraping years’ worth of data on one’s social media account is not only an invasion of privacy, but it chills freedom of expression.
AI’s ability to harness and process vast amounts of data from social media allows it to predict facts an applicant may have intentionally left out, such as age, race, religion, or political affiliation. Having acquired these facts, an employer can achieve systematic job discrimination in the thousands while hiding behind the “neutrality” of AI. It can use proxy factors as lawful excuses not to hire an applicant based on poor cultural fit. Most of all, employers can evade litigation simply because of the high burden of proof that a plaintiff needs to prove that discrimination was the ‘but for’ factor which resulted in him not getting the job.
When it comes to the accessibility of online data, such as public social media accounts, the U.S. legal system struggles with a binary concept of privacy in which it sees personal data as being private or public but never anything in between. While in the European Union (EU), the collection and processing of personal data is centered on consent and relevance for a specific purpose and not the manner on how data is made available.