Trauma behind the algorithm: a phenomenological analysis of violent content data work

Area of Science:

• AI Ethics
• Labor Studies
• Psychology

Background:

• Artificial intelligence (AI) development relies on extensive human labor for data processing and annotation.
• This 'ghost work' is often precarious, with poor conditions and significant burdens on workers.
• Violent Content Data Workers (VCDWs) are crucial for AI safety but face unique occupational hazards.

Purpose of the Study:

• To analyze the traumatic effects of reviewing violent content on VCDWs.
• To expose the hidden nature and ethical implications of this labor in AI.
• To highlight the unprecedented exposure levels faced by VCDWs compared to other professions.

Main Methods:

• Phenomenological analysis of VCDW experiences.
• Qualitative examination of the psychological and intersubjective impacts of the work.

Main Results:

• VCDWs experience profound trauma affecting affectivity, embodiment, and intersubjectivity.
• Unprecedented duration, velocity, and accumulation of exposure to violent content lead to compassion fatigue and deciphering difficulties.
• Workers suffer physical, psychological, and relational problems due to the nature of their work.

Conclusions:

• The work of VCDWs is systematically concealed through NDAs, outsourcing, and 'responsible AI' branding.
• The ethical frameworks for AI must account for the severe human cost of data annotation and content moderation.
• Urgent attention is needed to address the well-being and recognition of VCDWs.