Ethics and Artificial Intelligence: Accountability (7/7)
Principles for trustworthy artificial intelligence
In 2019, the European Commission's High-Level Expert Group on Artificial Intelligence (AI HLEG) published its "Ethics Guidelines for Trustworthy AI". To this day, it remains one of the foundational documents on the ethical requirements for artificial intelligence.
The Guidelines set out seven key requirements for trustworthy AI that must be observed at every stage of a system's lifecycle - from design and deployment to practical use in a real-world environment.
These requirements are:
-
Human agency and oversight - read our article here
-
Technical robustness and safety - read our article here
-
Data governance and privacy - read our article here
-
Transparency - read our article here
-
Diversity, non-discrimination and fairness - read our article here
-
Societal and environmental well-being - read our article here
-
Accountability
In this blog post we examine the last of the principles for trustworthy artificial intelligence, namely accountability.
What does the principle of accountability mean?
Accountability is a key principle that ties together all the other requirements for trustworthy artificial intelligence. It ensures that AI systems are not "black boxes," but are open to inspection, oversight and correction at every step of their lifecycle - from design and deployment to everyday use.
More specifically, the principle of accountability requires putting in place clear mechanisms that make it possible to trace who made a decision related to the AI system, when, and how.
Accountability encompasses the responsibility of AI system providers and deployers toward society, toward users, and before the law.
It must be applied:
-
Before a new AI system is introduced - through risk assessment, testing and prior documentation.
-
During its use - through ongoing monitoring, reporting and opportunities for correction.
-
After decisions have been made and actions taken - through the possibility of appeal, audit and legal redress.
Key elements of accountability
Auditing
AI systems should be subject to internal and external auditing of their algorithms, data and design processes. This is especially important when developing high-risk or specialized applications.
Assessment and documentation
Every decision should be accompanied by documentation of the data and algorithms used and the potential trade-offs involved. Tools such as artificial intelligence impact assessments help identify and minimize negative effects on individuals and society.
Legal redress
When a system makes an adverse or incorrect decision, users must have access to clear legal redress mechanisms. This includes the right to appeal and the right to opt out of automated decision-making.
Examples of accountability
1. Algorithmic auditing in New York (USA)
Since 2023, employers in New York that use automated tools for screening job candidates have been required to subject these systems to an independent bias audit. The auditors check whether the algorithms discriminate on the basis of sex, race or age. The results must be made public, which ensures accountability both to job applicants and to regulators.
2. The European Union - the AI Act and FRIA
The Artificial Intelligence Regulation (AI Act) requires that fundamental rights impact assessments (Fundamental Rights Impact Assessments, or FRIA) be carried out for high-risk AI systems. These include:
-
analysis of risks to rights such as non-discrimination, privacy and access to justice;
-
documentation of the objectives, data and algorithms;
-
measures to mitigate negative effects. These assessments are mandatory for systems that affect education, employment, migration and law enforcement.
3. Canada - the Directive on Automated Decision-Making
The Government of Canada requires all federal institutions that use AI for decision-making (for example in immigration and social services) to carry out Algorithmic Impact Assessments (AIA). This is a form of impact assessment that classifies risk from Level 1 (low) to Level 4 (very high). Depending on the risk, additional measures apply: independent auditing, public documentation and appeal mechanisms.
4. The Netherlands - SyRI and judicial review
The SyRI (System Risk Indication) system, used to detect welfare fraud, was declared incompatible with human rights by the court in The Hague in 2020. The main argument was the lack of sufficient accountability and transparency. This case underscored the importance of fundamental rights impact assessments and of independent oversight when deploying systems of this kind.
