Hold People Accountable for Your Technology And Its Impacts

October 5, 2021

This blog post is part of a series where weintroduce some basic concepts of responsible AI to SMEs developingAI based systems.

An essential part of an ethical and trustworthy AI is the concept of accountability. Accountability refers to the expectation that an entity or individual(s) ensure the proper functioning of the AI systems throughout the AI system’s lifecycle.

However, what is trustworthy and ethical AI in terms of accountability? Ethical and trustworthy AI is an AI that functions in accordance with the purpose it is designed and developed for. Moreover, ethical and trustworthy AI operates in conformity with relevant regulatory frameworks.[1]

In practical terms this means that in order to ensure that your AI is trustworthy and ethical, a set of mechanisms must be put in place to ensure accountabilities and responsibilities for the AI systems’ outcomes throughout its lifecycle, including for design, development, deployment and AI system use.[2] One such mechanism to ensure accountability is to assign responsibilities for the AI system, the assigned people being responsible for development and maintenance of the AI systems.

Even though assigning responsibilities is one of the mechanisms to ensure accountability, the concept of accountability refers, in general, to measures that illustrate governance on an organization-wide-level. There measure include, but are not limited to [3] [4]: 

  • Setting goals and objectives for the AI system; What does the system aim to achieve, what is it’s purpose?
  • Defining the responsible parties for the AI system; Who are the responsible organizations and departments? How can they be contacted?
  • Defining the responsibilities of the responsible parties; What are the responsible parties responsible for, how do the responsibilities and responsible persons change throughout the lifecycle of the system?
  • Auditability of the system; Is the system auditable and has it been audited?
  • Minimization of negative impacts and trade-offs; Have possible negative impacts of the system been considered and how have these impacts been mitigated? Are and possible trade-offs ethical and appropriate?
  • Reporting, contestability, remedies and redress; Are people to report negative impacts of the system or contest its outcomes? Is there a process in place to allow people to contest and is redress available?

Therefore, from the above, we can see that accountable AI refers to a system which operations are ensured and secured by setting up accountabilities and responsibilities related to the system. Furthermore, it is important to note that when responsibilities for the AI system are assigned, responsibilities of external actors, such as third party developers, should betaken into consideration as well. Therefore, when reviewing the people accountable for the AI system, also external parties that have key role either in the development or maintenance of the system, should be considered.

Lastly, an integral part of the accountability requirement is the proper documentation of the accountabilities. This is important not only for the teams developing and overseeing the systems but also for other stakeholder, whether external parties responsible for the AI system or individuals impacted by the systems or other relevant stakeholders. Documentation of the entity’s accountabilities ensures transparency, brings clarity to the responsibilities of the different actors, and provides stakeholders and impacted parties necessary information to ensure accountability in practice, such as how to contest a decision and who to contact about this.  

To conclude, an AI is only as trustworthy and transparent as the extent of its accountability-measures. An accountable AI ensures that AI operates in accordance with its purpose and in compliance with relevant regulatory instruments. An accountable AI is an AI that inspires trust.

[1] OECD AI Principle: Accountability.

[2] High-Level Expert Group on AI by European Commission: Ethics Guidelines for Trustworthy AI.

[3] European Commission: Artificial Intelligence Act. See also U.S Government Accountability Office: an Accountability Framework.

[4] High-Level Expert Group on AI by European Commission: Ethics Guidelines for Trustworthy AI and High-Level Expert Group on AI by European Commission: Assessment List for Trustworthy Artificial Intelligence (ALTAI).

Collaborate with us

We collaborate with partners who work in different ways to create a better future with AI. Connect with us to talk about how our platform can help your and your customers' businesses thrive.