Microsoft Sued For Using Image Dataset To Improve Its Facial Recognition Technology


Plaintiffs Steven Vance and Tim Janecyk filed a class action complaint against Microsoft on Tuesday in the Western District of Washington for its use of their photographs in an image dataset obtained by Microsoft, which the company used to improve its facial recognition technology.

The plaintiffs proffered that “[i]n its effort to improve its facial recognition technology, Defendant Microsoft violated Illinois’ Biometric Information Privacy Act 740 ILCS 14/1, et seq. (“BIPA”), by, among other things, unlawfully collecting, obtaining, storing, using, possessing and profiting from the biometric identifiers and information of Plaintiffs” and the class. BIPA seeks to protect individuals’ biometrics, or unique features that are used to identify an individual, such as fingerprints and facial geometry. The plaintiffs added that facial biometrics are easy to observe online, consequently, this poses a privacy threat to individuals.

In order to collect or obtain biometrics, BIPA requires entities to “provid[e] written notice and [obtain]…a written release”; entities are “prohibited from profiting from an individual’s” biometrics; and entities are required “to develop a written release policy, made available to the public, that establishes a retention schedule and guidelines for permanently destroying such identifiers and information” if entities are in possession of biometrics.

Microsoft “developed, produced, marketed and otherwise used facial recognition products and technologies,” specifically its Cognitive Services Face Application Program Interface and its Face Artificial Intelligence service products, which allowed “customers to embed facial recognition into their apps without having to have any machine learning expertise.” Microsoft sought to improve its technology.

Photo sharing website Flickr allegedly “had access to over 100 million photographs posted” by its users; these images were compiled into a single publicly available dataset. However, this was done without informing individuals or obtaining their consent. Flickr stated it made the dataset public to improve the accuracy and reliability of facial recognition technology, which is inaccurate especially when identifying women or people of color. A 2018 study indicated that a Microsoft facial recognition product was more inaccurate when identifying women and people of color. In response to the study, IBM used the Flickr dataset to create a Diversity in Faces Dataset, using “19 facial landmark points from each image in the dataset to determine 68 key points for each face.” Microsoft obtained the Diversity in Faces Dataset from IBM. As a result, “Microsoft used the links provided by IBM to download, copy or otherwise obtain from Flickr each photograph in the dataset…in order to associate the biometric identifiers and information provided by IBM with the actual photographs to which the biometric data related.”

However, the plaintiffs, whose images are supposedly in both the Flickr and Diversity in Faces Dataset, claimed that Microsoft violated BIPA because the individuals whose images were used in the Diversity in Faces Dataset were not notified and Microsoft via IBM did not obtain permission and the required written release to use their photographs. Furthermore, the plaintiffs alleged that Microsoft profited from the biometrics in the Diversity in Faces Dataset because “those biometric identifiers and information allowed Microsoft to improve its facial recognition products and technologies…by allowing Microsoft to improve the effectiveness of its facial recognition technology on a diverse array of faces, thereby making those products and technologies more valuable in the commercial marketplace.” These facial recognition products were then sold to clients, thus also financially benefitting Microsoft. Additionally, the plaintiffs asserted that their privacy has been violated from this conduct. Microsoft is also accused of unjust enrichment for its alleged conduct.  IBM and Flickr are not named as parties to the lawsuit. 

The plaintiffs, represented by Loevy & Loevy as well as Carlson Lynch LLP, have sought to certify the class and for plaintiffs and their counsel to represent the class; an award for damages, including statutory and punitive damages; equitable, injunctive and declaratory relief; a disgorgement of unjust enrichment and proceeds; pre- and post-judgment interest; an award for costs and fees; and other relief as determined by the court.