Human dignity
A people-first assessment should examine whether AI serves those affected rather than treating them merely as inputs to an optimisation process.
Responsible AI practice IWIB4AI session Needs review
Workshop – Responsible AI Self-Assessment: Start with Question Zero
Responsible AI starts by asking whether AI should be used, before deciding how to use it.
Draft – legal and case law references not yet verified
Summary
This use case concerns a Responsible AI Self-Assessment Tool for public, private and academic organisations. Its people-first starting point is “Question Zero”: whether AI should be used at all, before considering how to implement it. This approach puts the justification for AI adoption ahead of technical implementation.
The Problem
Organisations may treat AI adoption as a given and focus on managing implementation risks rather than questioning whether it serves people's needs. The core challenge is deciding whether AI is justified, including whether a non-AI approach could achieve the objective with fewer harms.
Ethical Grounds
Autonomy
The initial decision should consider whether using AI would strengthen or constrain people's ability to make meaningful choices, including whether realistic alternatives remain available.
Fairness
Assessing whether AI is appropriate requires examining who benefits and who bears its risks, including whether a non-AI alternative would avoid unequal burdens.
Accountability
Organisations should identify who is responsible for the adoption decision and document its justification. Completing a self-assessment should support, not replace, responsibility for foreseeable impacts.
Legal Compliance
| Framework | Provision | Relevance |
|---|---|---|
| EU AI Act | Art. 9 | For providers of high-risk AI systems, self-assessment can inform—but does not replace—the required ongoing risk-management system. |
| EU AI Act | Art. 27 | Deployers covered by this provision must conduct a fundamental rights impact assessment before deploying specified high-risk AI systems; a general responsible-AI self-assessment is not automatically sufficient. |
| GDPR | Art. 35 | Where intended personal-data processing is likely to create a high risk to individuals' rights and freedoms, a data protection impact assessment must address necessity, proportionality, risks and safeguards before processing begins. |
Sources
Court & Regulator Decisions
NJCM et al. v State of the Netherlands (SyRI) (opens in a new tab)
The invalidation of welfare-fraud risk profiling illustrates why public-interest objectives do not remove the need to assess privacy impacts and safeguards before adoption.
IMY – Skellefteå municipality facial recognition in school (opens in a new tab)
Unlawful facial recognition for attendance demonstrates why self-assessment should question whether intrusive AI is justified and whether consent is meaningful.
Key Takeaways
- Define the human need before selecting an AI solution.
- Compare AI with non-AI options, including not proceeding.
- Document affected groups, expected benefits, potential harms and the accountable decision-maker.
- Use self-assessment to inform applicable legal assessments, not as proof of compliance.
Related Use Cases
From the Session
Workshop – Responsible AI Self-Assessment: Start with Question Zero
Speakers
