Fairness
AI should be assessed for how it distributes benefits, costs and opportunities across communities and countries, not merely for aggregate performance. Existing inequalities can worsen even where a system appears technically accurate.
Democracy & information IWIB4AI session Needs review
Race to the Bottom: AI in a New Age
Addressing how AI can concentrate power, deepen global inequality and weaken democratic institutions.
Draft – legal and case law references not yet verified
Summary
This use case examines how AI development and deployment can reinforce existing power structures, leaving less powerful communities with fewer benefits and greater risks. It focuses on democratic institutions’ vulnerabilities and disproportionate impacts on the Global South. Technical safeguards, public policy and civil-society participation offer complementary ways to address these imbalances.
The Problem
Those controlling AI infrastructure and deployment may capture its benefits while shifting social and democratic risks onto people with limited influence over its design or use. How can AI governance redistribute decision-making power and ensure that affected communities can challenge harmful outcomes?
Ethical Grounds
Democratic integrity
Concentrated control over AI can amplify private influence over public debate and institutional decisions. Safeguards should protect democratic participation and institutions’ capacity to scrutinise powerful actors.
Inclusion
Communities in the Global South and other underrepresented groups should have meaningful influence over AI priorities and safeguards. Consultation must account for differences in language, resources and access to decision-makers.
Accountability
AI providers and deployers should remain answerable for societal impacts rather than treating harm as an unavoidable consequence of innovation. Independent scrutiny and accessible routes to challenge decisions help civil society hold them accountable.
Legal Compliance
| Framework | Provision | Relevance |
|---|---|---|
| EU AI Act | Art. 27 | Specified deployers of certain high-risk AI systems must assess fundamental-rights impacts, providing a mechanism to identify harms to affected groups before deployment. |
| DSA | Arts. 34–35 | Very large online platforms and search engines must assess and mitigate systemic risks, including risks to civic discourse, electoral processes and fundamental rights arising from their services and algorithmic systems. |
| GDPR | Art. 35 | AI processing of personal data likely to create high risks to individuals’ rights and freedoms requires a data protection impact assessment, although this does not replace broader societal-impact analysis. |
| EU Charter of Fundamental Rights | Arts. 21 and 51 | The prohibition of discrimination constrains EU institutions and Member States when implementing EU law, including relevant public decisions concerning AI governance and use. |
Sources
Court & Regulator Decisions
Annulment of the 2024 Romanian presidential election Needs review
The election annulment illustrates democratic institutions’ vulnerability to coordinated online manipulation, informing governance of AI-amplified political influence.
Google and Alphabet v Commission (Google Shopping) (opens in a new tab)
The decision illustrates how dominant platforms can entrench power through self-preferencing, informing scrutiny of concentrated AI distribution and infrastructure.
NJCM et al. v State of the Netherlands (SyRI) (opens in a new tab)
The successful challenge to opaque welfare-fraud scoring demonstrates how civil society can contest government algorithms that impose risks on less powerful communities.
Key Takeaways
- Map who controls an AI system, who benefits and who bears its risks.
- Evaluate distributional and democratic impacts alongside technical performance.
- Give affected communities, including those in the Global South, meaningful roles in design and oversight.
- Combine technical safeguards with independent scrutiny and accessible ways to challenge harmful outcomes.
Related Use Cases
From the Session
Use Case #2 (2025) – Race to the Bottom: AI in a New Age
Speakers
