Algorithmic Audits
Adversarial Algorithmic Auditing Guide
Algorithmic auditing is an instrument for dynamic appraisal and inspection of AI systems. This guide focuses on adversarial or third-party algorithmic audits, where independent auditors or communities thoroughly examine the functioning and impact of an algorithmic system, when access to the system is restricted.
Algorithmic Audit · Social Impact · AI Ethics ·
Adversarial audits bridge the gap between innovation potential and societal impact.
The proposed approach is systematic and adaptable, incorporating qualitative contextual analysis, stakeholder mapping, evaluation of bias and inefficiencies, and reverse engineering of system processes through research.
Adversarial audits become particularly significant when organizations are unwilling to undergo internal audits or when regulatory requirements for algorithmic audits are lacking.
By serving as an independent mechanism to uncover potential negative impacts of algorithms and AI systems, these audits prompt developers to address issues and inform regulators and the public for appropriate actions. The ultimate goal is to ensure that AI systems operate with fairness and accountability, minimizing harmful consequences for society and protected groups.
Case Studies
Our Adversarial Audits
VioGén, determining the risk of a gender-based violence victim
VioGén is an algorithm that determines the level of risk faced by a victim of gender-based violence and establishes her protection measures in Spain. It is the largest risk assessment system in the world, with more than 3 million registered cases. It is designed as a recommendation system but, even though the police officers are able to increase the automatically assigned risk score, they maintain it in 95% of the cases.
Therefore, it is the system that decides which police protection measures are assigned to each victim. In 45% of cases, that level of risk assigned is “unappreciated”.
Consequences are clear, between 2003 and 2021 there were 71 murdered women who had previously filed a report without obtaining a level of risk that entails police protection. The system only gives the number of “extreme” risk scores (which means risk of homicide) it can afford, so funding cuts have a direct and quantifiable impact on the chances that women will have in order to receive effective protection after seeking police protection.
In terms of transparency, most VioGén studies have been conducted by the same researchers that contributed to its development. This reinforced the argument for the need of an independent oversight of the system.
YouTube and TikTok and the representation of migrants
TikTok wants to entertain, not talk about politics.
Being one of the most popular and influential social media platforms, TikTok can also affect the political discourse on migration. However, the audit found that TikTok’s recommender system shows little variation in recommended content regardless of users’ attitudes towards migration and their location’s political leaning.
The platform’s weak personalization for political content indicates that TikTok’s focus is primarily on entertainment rather than politics.
YouTube’s algorithm reinforces a dehumanizing view of migrants.
Migrants and refugees are often depicted in a negative light on this platform, predominantly as non-white individuals crossing borders. This perpetuates their dehumanization and stereotyping, which can have real-world consequences on how they are perceived and treated by society.
Uber, Cabify, and Bolt
Ride-hailing apps like Uber, Cabify, and Bolt use surge pricing algorithms to determine ride fares based on supply and demand in a given area and time, resulting in geographic price discrimination. There are indications that prices in ride-hailing apps tend to be lower in more affluent neighborhoods.
This raises concerns about algorithmic price discrimination on the basis of the geographic location and socio-economic makeup of neighborhoods, making mobility services less accessible to disadvantaged groups.
Competition law exists to protect consumers and ensure they have true choice. It also indirectly benefits businesses, the public sector, and the economy as a whole.
The report findings suggest that the pricing algorithms of Uber, Cabify, and Bolt are colluding in some of the most important routes in Andalusia and Madrid, breaching Law 15/2007 for the Defense of Competition (LDC) in Spain. This harms other actors in the market, such as traditional taxis and potential new entrants.
Adversarial Audits scope
Independent Audits required by the DSA: A matter of compliance
The Digital Services Act will require digital services companies, including social networks, to conduct independent audits and risk assessments to ensure a safe digital space in which the fundamental rights of users are protected.
While independent audits are carried out upon the request of the company implementing the system, which is a service Eticas provides as well, adversarial audits play a critical role in raising awareness about some of the systems present in our daily lives that have a quiet yet enormous impact.
