Lessons From FTC Action Against Rite-Aid's Facial Recognition
A major FTC ruling against Rite Aid for "recklessly" bad facial recognition illustrates important best and worst practices for deploying the technology.
Face recognition accuracy remains a problem despite significant improvements e.g., Facial Recognition Improvements Wanted 2023, Facial Recognition Rankings 2023.
In this post, IPVM examines the major lessons learned from the FTC's case.
Executive *******
*** *********** ***'* ****-********** ****** *********** ****** "thousands" ** ********** **** **** ** 2020, ******* **** *** **** ***** the **** *** * *****.
*** ***'***************** *** *** **********, **** ** images ******** ** ****, *******, *** poor ********/******, *** **** ***** *********. The **** **** ****** **** ********* for **** * "****" ****** ***** look ****, *********:
- ********* ****** **** ****** *********** ** being ****.
- ****** **** *** ***** ** ******* their **********.
- ********** ******* ** *** ********** ******.
- ****** ******* ** ****** **** ** gender ******.
*** **** **** *** ******** ****** other ****** *********** ***** ** ** is ***** ** **** *** ********* an***** *** ****** ******* **. **** *** ******** *** ***'* decision, ****** ** ***** **** *********** 3 ***** *** ******. *** ****** was ******* ** *******, * ***** PRC/US *******, ********* ** ***** ******* *************.
FTC: **** *** "****** ** ******* ****"
*** ************** ******** ** **** **** *** "failed ** ********* ********** ********** *** prevent **** ** ********* ** *** use ** ****** *********** ********** ** hundreds ** ******" **** **** ** 2020, ******* **** *** **** ********* the **** *** * ***** *** forcing ** ** ****** *** ****** images ********.
**** *******"** ************* ********" **** *** ***'* complaint (**** *** *********) *** ******** the *******, ****** ** **** **** facial *********** * ***** *** ****** following ******** *************. *******, ** **** ********* **** issued.
Direct ****** ******* ** **** ***
********* ****** ******, *** ** *** ** federal *** ********** ****** ***********. *** basis ** *** ***'* ****** *** not **** *** ********* *** ******** laws, *** ********* ***** *** ******** ****** ******* ******** ****.
FTC ****** "********" **** *********
*** ***'* ********* ******* **** *** create "* ******** *** **** * comprehensive *********** ******** ******* ****** **** like," **** *********** *** ****************** ******. ***** ******** *******:
- ******* "******* ******" ** ****** **** they *** ******** ** * ****** recognition ****** *** **** **** *** to ******* ***** ***** (*.*. *** a ********* ****).
- ** ****** ****** ***** ** * match, ****** ****** **** ******** *** how ** *******.
- ******* ** ********** "*******," *.*. ******* receipt ****** * **** *** ******* officially ****** ** ****.
- ******* *********** ******** *** **** *********** annually.
- **** **** *** ****** ** ** cannot ******* *** ***** **********.
*** ***'* ****** **** ********* "** one ****** **** **** **** **** settlement" ******** *** *** "************* ******** the *** ** ********* ************ ** commercial ********".
Mismatches ****** ** **** ***** *******
**** *** "****** ** **** **** of *** **** ***** ***********" **** cameras ***** ********* ********** *** ** "blurry" ********, *** *** ***'**********:
*** *** **** *** ******* *** they **** ******, *** **** ** often *** **** ** *** ******* are *** **** ** ** *** away. *******, **** "********** ******" ** suspected *********** *** "**** ********" ********* "overexposure, *****, *** *** ******* *****":
*** *** ***** **** **** ********** images ******** ** ***** **** "****** are *** *******". *******, **** *** enrolled **** ******** ** ****** **** hats *** ******* *** *** ******, leading ** **** **********:
*******, **** ***** **** ********* *** were *** ***** ********** ****** ** gauge *** ******** * '*****' ****** was.
How ****-***'* ****** ****** *********** ******
**** *** ***** "******" ****** ****** of ********* *********** *** **** ******* those ** **** ******* ** ****** entering ******** ** *** ******. ** there *** * '*****', ***** ***** would ******* ** ***** ** ***** smartphones **** **** ****** ****-**-****.
*** ****** *** ****** *.*. *** disclosed ** *********. ******, ******* ** *******, * ******* "**** ** *********" sign *** **** *******:
** ********, ******* ***********, *** **'* **** ****-********** ****** recognition *******, ***** **** *** ****** to ******** ** *** ***** ****:
(*********'* ******* **** **** *** ***** obscured ***** ** * **** ****/**** on *******************).
Real-Life **** ********
*** ********************** ******** ** ****-**** ***** ****** by **** ***'* ******, ********* ** 11-year-old **** ***** *******, ******* *** mother ** **** ****:
* **** *** ******** ******* *** searched ** **-****-*** **** ******* ** a ***** *****. *** ****’* ****** reported **** *** ****** **** ******* her ******** *** ** ********** ***** the ********.
** ******* *******, * ***** ****** was ******* ** * '**********' ******* the ****** ******* ***** * ***** woman **** ****** ****:
**** *** ********* ****** *** ****** on * ******** ******* *** ********** generated ** ***** ******* ** ***** that *** ***** ********* ** ********* “a ***** **** **** ****** ****.” The ******** *** *****.
*******, ******** ***** *** ****** "******* customers **** ****** *** *** ********** been ******** ** *** ******** ***** on ******** ********* ** ***** ****".
System ******** ******* **********
**** ***'* ****** *********** ****** *** powered (**** ******** *** ********) ** DeepCam, * ***** ***/** ******* ***** IPVM*********** ******* $** ****** *********** ******* at *** **** ** ****. *** cameras **** **** *******'****** ********, *** *** ******** ********************.
******* **** ******* ** **** *** US ****** ** ****. *******'* ******** co-founder,*** ******, *** ************, * ****** *********** ******** *** US *** ***********, ***** ** **** Reuters ** '*** * **********' ** DeepCam.
Facial *********** ******** ***** **********
******** *** ****** *********** ** ***** surveillance *** ******** ************* ***** **** Aid's ****** *** **** **** **** to ****, *** ** ******* * major *****, ** **** ******** ******** *********** ************ ****** ************* *********** ******** ****.
**** ***'* **** ******* ******* ** earlier *** **** ********* ****** ***** gave ***** *********** ** *************. **** false ****** ******** ***** **** ***, e.g.********* "**%+ ****** *********** ********" ******.************** ** **** *****, ***** ** better *** **** ********* ** *** difficult *** ***** ***** **** ****** recognition **.
Showcases **** ** **** ******** / ********
*** ***'* **** ***** *** ***** of ********* **** ****** *********** ******* strong ************ *********, ***** ********, *** clear ******** *** ******** **********.
*** ***'* ***** **** **** ****** demand *** ****-********** ****** *********** ** the **, **** *** ***** ********** to ***** *** ******* ******** *** the *********** *****/************ ******** **** ******* up ***** ********* ******* ** *** first *****.
**** ** * ****** ******** ** the **, ***** *** *******'* **** regulator********** *******'* ******* ********, *********, **** year.
*'* *** ***** **** **** *** a ******** ***** **** **** ******** else.
**** ****-******* ***** ***** **** ** difficult ** ******* * ****** *********** system **** ************ *** ***** ******* / ********* ** **********, **?
******* ***'** ***** **** ******** *** a ***** *****. *** *** *** blames **** *** ********. ** ***** in*** *********, **** *** *** *** ****** if *** ***** ********* ********:
******** ** *** **** ***’* ****** that *** ****** ****** ******* ********* authorized ** ******* ****** *********** ********** withapproximately *** ** *** ***** ** ******** on its facial recognition system, in nearly all cases Rite Aid did *** ****** ** ****** *** ****** that employees had received the required training. [emphasis added]
*******, *** ******** *** *** ******** address *** **** ** ***** *********:
********, **** ***’* ******** ********* **** very ******* *** *** *** ******* the ***** ** ********* **** ***** the **********. ** ******** *********, *** training ********* **** *** ******** ** train *****-***** ********* *** ******** **** Aid’s ****** *********** ********** **** ******* to ****** **** ** *** ** navigate *** ******** *** ****** ************ used ** ******** **** *** **********, and *** ** ***** *** ***********. Rite ***’* ******** ********* ******did *** ******* *** *********** **** *** ********** ***** ******** *****-******** ***** ****** or contained only * ******* ********* to such a possibility. [emphasis added]
*** *** **** *** ******* *******, the ******'* ********, *** *** **** in *** *******.
**** ****-******* ***** ***** **** ** difficult ** ******* * ****** *********** system **** ************ *** ***** ******* / ********* ** **********, **?
** ***** **** ** ********* ** you **** **. **** ** *** the *** ****** ** ***** ******** of ***** ********** ** ***** ******** the ********.
** ******* ** ****** ******:
[** *** ****** ** ******** ***** you **** ****** ** *** ******* the ****** **** ** * *****. Does *** **** ** *** ******* exactly ***** *** ****** ***'** ******* at ****** *** ***** * ******* of *******? ***- ******* **** ****** policy. **- ********* *** **** ** with ****.]
*'* *** ***** **** **** *** a ******** ***** **** **** ******** else. ****** *********** ** *** * lot ********* **** ****** ******** ** the **** ******** ** ********* *** previously ********** ** ****** *** ******, which ********** **** **** *** *******. Facial *********** ** **** ** ********* way ** ***** ** ******* ***** people ** * ****** ******** ** pictures *** ******** ** *****.
*** ****** ******** ***** **** ******** employees **** ****** *********** ** *** from ******* *** **** **** ********, so ******** * **** ***** ***** to ******* *** ***** ***** ******* in **** *** ****** ***** **** been *****. ** ** *** ******* doesn't **** **** *** ****** ** the ******* ** ***, **** ****** it ***'*.Use ****** ***** **********.
** ***** *** *** **** ** improper ******** **** ****? *** *** product *** ******** *** ********** *** expectations? ** *** **** **** ****** disclosures, *** **** *** ********** ****'* listen, ****** ******* *** **** ****** that ******** ****** ************? *** **** fail ** ****** ********* ** *** discretion?