From Oppression to Empowerment: Regulating AI

Core argument:

• "No AI system is ever neutral" (Floridi 2023; Wajcman 2010)
• AI amplifies existing inequities (Toyama 2011)
• AI is reinforcing oppression, negatively affecting marginalized people while the already privileged benefit most

3 research pillars (transformation potentials):

1. Regulating AI — e.g., EU AI Act implementation

2. Governing AI — responsible AI governance strategies

3. Reimagining AI — feminist-informed redesign

Gender Data Gap:

• Systematic gaps, distortion, or invisibility of gender in data — who is recorded, how data is recorded, what data is used for
• Can have negative impacts on women (e.g., medical devices calibrated on male bodies)

The socio-technical entanglement:

• AI outcomes (predictions, recommendations, generations) reflect and reinforce societal bias
• AI and societal systems are "entangled" — the world as it IS ≠ the world as it SHOULD be
• Data reflects existing world → model inherits existing biases → perpetuates them

Power-over vs. Power-to:

• Current AI operates via 'power-over' dynamics (interlocking systems of oppression: patriarchy, colonialism, capitalism)
• Goal: transform AI toward 'power-to' dynamics (equity, democracy, sovereignty)

Key concepts from intersectional feminism:

• Intersectionality (Crenshaw 1989): multiple identity dimensions (race, gender, class) interact to create overlapping discrimination
• Matrix of domination (Collins 2000): structural, hegemonic, disciplinary, and interpersonal domains of power
• LLMs reflect the ideology of their creators (Buyl et al. 2024)

Real-world examples of oppressive AI:

• Amazon's AI recruiting tool was scrapped because it showed bias against women (Dastin 2018)
• Predictive policing algorithms remain racially biased regardless of input data
• Facial recognition systems have wrongly caused arrests due to misidentification
• Iran using drones and phone apps to monitor dress code for women

Practical implications (5 target groups):

1. Regulators: don't relinquish AI regulation; design mandatory diversity requirements

2. Organizational leadership: practice responsible AI governance; aim beyond mere compliance

3. Development teams: practice reflexivity; implement feminist reflexes

4. Users/civil society: use AI for feminist purposes; adopt fair AI prompting strategies

5. Researchers: make positionality explicit; recognize knowledge is produced from situated standpoints