Incident: Passport Facial Recognition System Fails with Dark Skin Users

Published Date: 2019-10-09

Postmortem Analysis
Timeline 1. The software failure incident of passport facial recognition checks failing to work with dark skin happened in 2019. [90522]
System 1. Facial recognition technology used in the passport checking service [90522]
Responsible Organization 1. The Home Office [90522]
Impacted Organization 1. Individuals with very light or very dark skin were impacted by the software failure incident [90522].
Software Causes 1. The facial recognition technology failed to recognize the features of some people with very light or dark skin, indicating a software issue in handling different skin tones [90522]. 2. The system had trouble mapping the faces of some ethnic minorities, showing a software limitation in accurately capturing diverse facial features [90522].
Non-software Causes 1. Lack of proper testing and validation of the facial recognition technology with a diverse range of ethnic groups before deployment [90522]. 2. Decision to launch the service despite knowing about the system's trouble mapping faces of some ethnic minorities [90522]. 3. Design flaw in the facial recognition technology that led to difficulties in recognizing features of individuals with very light or very dark skin [90522].
Impacts 1. The software failure incident led to the facial recognition technology failing to recognize the features of some people with very light or dark skin, impacting their ability to use the passport checking service [90522]. 2. Users had to override the automated check if needed, indicating a disruption in the intended automated process [90522]. 3. The incident highlighted the difficulty faced by some ethnic minorities in providing an acceptable passport photograph, affecting their user experience with the service [90522]. 4. The rejection of a young black man's photograph due to the software mistaking his lips for an open mouth showcased a specific impact on individuals' applications being rejected based on erroneous facial recognition results [90522].
Preventions 1. Conducting thorough and inclusive user research with a diverse range of ethnic groups to identify potential issues with facial recognition technology [90522]. 2. Implementing rigorous testing procedures specifically designed to detect and address issues related to recognizing features of individuals with very light or very dark skin tones [90522]. 3. Delaying the launch of the passport checking service until the identified problems with facial recognition technology were adequately resolved [90522].
Fixes 1. Conduct thorough user research and testing with diverse ethnic groups to ensure the facial recognition technology can accurately map features of individuals with very light or very dark skin [90522]. 2. Implement improvements in the facial recognition algorithm to better handle a wider range of skin tones and features [90522]. 3. Consider integrating other facial recognition and tracking systems that are being developed to address the specific problem of recognizing features of individuals with very light or very dark skin [90522].
References 1. New Scientist 2. Freedom of Information request 3. Evening Standard [90522]

Software Taxonomy of Faults

Category Option Rationale
Recurring one_organization (a) The software failure incident related to facial recognition technology failing to work with dark skin has happened again at the Passport Office. The article mentions that this is not the first time the Passport Office's use of facial recognition has been problematic, as a young black man's photograph was recently rejected due to the system mistaking his lips for an open mouth [90522]. (b) The article does not provide specific information about similar incidents happening at other organizations or with their products and services.
Phase (Design/Operation) design, operation (a) The software failure incident related to the design phase is evident in the article. The Home Office's new passport checking service faced issues with facial recognition technology failing to recognize the features of some people with very light or dark skin. The problem was identified during trials of the service, indicating a failure in the design phase where the system had trouble mapping faces of some ethnic minorities despite user research acknowledging the difficulty faced by individuals with very light or very dark skin [90522]. (b) The software failure incident related to the operation phase is also highlighted in the article. The Home Office decided to launch the service despite being aware of the system's trouble in mapping faces of some ethnic minorities. Users were allowed to override the automated check if needed, indicating that operational procedures were in place to address the issue during system operation [90522].
Boundary (Internal/External) within_system (a) within_system: The software failure incident related to the passport facial recognition checks failing to work with dark skin was primarily due to factors originating from within the system. The facial recognition technology used by the Home Office had trouble recognizing the features of individuals with very light or dark skin, as revealed in the New Scientist report [90522]. The system's inability to accurately map faces of some ethnic minorities, including those with very light or very dark skin, was acknowledged in documents released as part of a freedom of information request. Despite being aware of these issues, the Home Office decided to launch the service, indicating that the failure was primarily within the system's design and functionality.
Nature (Human/Non-human) non-human_actions (a) The software failure incident in the article was related to non-human_actions. The facial recognition technology used in the passport checking service failed to recognize the features of some people with very light or dark skin, indicating a failure in the system's ability to accurately map faces from computer-scanned images or live feeds. This issue was identified during trials of the service, and the Home Office was aware of the system's difficulty in mapping faces of some ethnic minorities. The failure was attributed to the system's inability to handle certain shades of skin, which are non-human factors affecting the software's performance [90522]. (b) The article does not provide specific information about the software failure incident being directly caused by human actions.
Dimension (Hardware/Software) hardware, software (a) The software failure incident related to hardware: - The article mentions that the facial recognition technology used for passport checking had trouble handling some shades of skin, particularly very light or very dark skin. This indicates that the failure was due to the hardware's inability to accurately map the features of individuals with certain skin tones [90522]. (b) The software failure incident related to software: - The article highlights that the facial recognition technology failed to recognize the features of some people with very light or dark skin. This failure was attributed to the software's inability to properly map the faces of certain ethnic minorities, indicating a software-related issue [90522].
Objective (Malicious/Non-malicious) non-malicious (a) The software failure incident related to the passport facial recognition checks failing to work with dark skin does not seem to be malicious. The issue was due to the technology having trouble recognizing features of individuals with very light or dark skin, as reported in the New Scientist [90522]. The Home Office was aware of the problem but decided to launch the service regardless, indicating that there was no malicious intent behind the failure.
Intent (Poor/Accidental Decisions) poor_decisions (a) The intent of the software failure incident: The software failure incident related to the passport facial recognition checks failing to work with dark skin can be attributed to poor decisions made by the Home Office. Despite knowing that the facial recognition technology had trouble recognizing features of individuals with very light or dark skin, the Home Office decided to launch the service regardless. This decision was based on the judgment that the overall performance was sufficient to deploy, even though the system had trouble mapping faces of some ethnic minorities [90522].
Capability (Incompetence/Accidental) development_incompetence, accidental (a) The software failure incident related to development incompetence is evident in the article. The Home Office's new passport checking service faced issues with facial recognition technology failing to recognize the features of some people with very light or dark skin. Despite knowing about the problem during trials and acknowledging the difficulty faced by some ethnic minorities, the Home Office decided to launch the service regardless, stating that the overall performance was judged sufficient to deploy. This decision to proceed with a system that had trouble mapping faces of certain ethnic groups despite being aware of the issue reflects a level of development incompetence [90522]. (b) The software failure incident related to accidental factors is also present in the article. The article mentions a specific case where the photograph of a young black man was rejected because the system mistook his lips for an open mouth. This indicates an accidental failure in the facial recognition system, where the software made an incorrect interpretation leading to the rejection of a valid photo [90522].
Duration temporary The software failure incident described in the articles appears to be temporary. The facial recognition technology used in the passport checking service failed to recognize the features of some people with very light or dark skin during trials [90522]. The Home Office was aware of the problem but decided to launch the service regardless, indicating that the failure was due to contributing factors introduced by certain circumstances but not all. Additionally, the Home Office mentioned that it would try to improve the system, suggesting that the failure was not permanent.
Behaviour omission, value, other (a) crash: The software failure incident related to the passport facial recognition checks did not involve a crash where the system loses state and does not perform any of its intended functions [90522]. (b) omission: The software failure incident could be related to omission as the facial recognition technology failed to recognize the features of some people with very light or dark skin, indicating an omission in performing its intended functions for certain individuals [90522]. (c) timing: The software failure incident was not related to timing issues where the system performs its intended functions correctly but too late or too early [90522]. (d) value: The software failure incident could be related to a value issue as the facial recognition technology was not performing its intended functions correctly by failing to recognize features based on skin color, leading to incorrect performance for certain individuals [90522]. (e) byzantine: The software failure incident was not related to a byzantine behavior where the system behaves erroneously with inconsistent responses and interactions [90522]. (f) other: The software failure incident involved a specific issue where the system had trouble mapping the faces of some ethnic minorities, particularly those with very light or very dark skin, which led to difficulties in providing acceptable passport photographs [90522].

IoT System Layer

Layer Option Rationale
Perception None None
Communication None None
Application None None

Other Details

Category Option Rationale
Consequence unknown (a) death: People lost their lives due to the software failure (b) harm: People were physically harmed due to the software failure (c) basic: People's access to food or shelter was impacted because of the software failure (d) property: People's material goods, money, or data was impacted due to the software failure (e) delay: People had to postpone an activity due to the software failure (f) non-human: Non-human entities were impacted due to the software failure (g) no_consequence: There were no real observed consequences of the software failure (h) theoretical_consequence: There were potential consequences discussed of the software failure that did not occur (i) other: Was there consequence(s) of the software failure not described in the (a to h) options? What is the other consequence(s)? The articles do not mention any direct consequences such as death, harm, impact on basic needs, property loss, or non-human entities being affected due to the software failure. It seems that the main consequence discussed is the issue of the facial recognition technology failing to work properly with certain shades of skin, particularly very light or very dark skin, which led to difficulties in providing acceptable passport photographs [90522].
Domain government The software failure incident described in the articles is related to the government industry. The failed system was a passport checking service launched by the Home Office, which encountered issues with facial recognition technology not being able to recognize features of individuals with very light or dark skin [90522]. The government was aware of the problem but decided to launch the service regardless, stating that user research had identified difficulties faced by individuals with very light or very dark skin in providing acceptable passport photographs [90522].

Sources

Back to List