Introduction to AI 2
by Albert Major
Case 1: Safe Spaces vs. Exploitative AI Labor
Topic: AI relies on human labor to filter content, but the conditions of these workers are questionable.
- AI models like ChatGPT need human moderation to filter harmful content.
- Example: 200 Kenyan workers were paid very little to filter toxic content for OpenAI, then fired.
- Content moderation jobs often involve seeing horrific, violent, or disturbing material.
- Workers were paid $1.32 to $2 per hour, much lower than wages in developed countries.
- Many AI companies, including Meta and OpenAI, have faced criticism for outsourcing content moderation to low-wage countries.
Debate: Should AI content moderation be outsourced this way?
- Pro (Content Moderation is Necessary)
- AI models must be trained with human feedback to avoid harmful, racist, sexist, or violent responses.
- Without moderation, AI could become extremely biased or used for illegal purposes (e.g., building weapons).
- CAPTCHAs and other tasks show how human labor is already a key part of AI development.
- Con (Ethical Concerns)
- The mental health impact on moderators is severe—they are exposed to harmful content daily.
- AI companies exploit low-cost labor while keeping the benefits.
- This raises ethical concerns about worker exploitation vs. user safety.
Philosophical Questions Raised
- Is this a modern version of the “trolley problem”?
- AI companies “pull the lever”: They sacrifice a small group of workers (Kenyan moderators) to protect millions of users from harmful AI-generated content.
- Is it ethical? Is it truly “less harm” overall?
- Is AI moderation censorship?
- Some argue users should decide what to see and AI should be neutral.
- Others argue AI is not a human and should not have free speech rights.
Case 2: Hidden AI Labor (Amazon Mechanical Turk & AI Data Annotation)
Topic: Many AI tasks that seem automated are actually done by underpaid human workers.
- AI companies use platforms like Amazon Mechanical Turk to train AI models with human labor.