These Women Tried to Warn Us About AI
ai-ethicsalgorithmic-biasfacial-recognitionai-safetywomen-in-aitech-criticism
Abstraction: Women researchers warned of AI bias harms
Key points:
- Timnit Gebru co-led Google's Ethical AI team and co-authored the "Stochastic Parrots" paper on LLM dangers; she was fired/resigned in late 2020 after Google asked authors to remove their names.
- GPT-2 examples of bias: "the woman worked as" → "a prostitute"; "the Black man worked as" → "a pimp"; training data skewed (Wikipedia <15% women contributors, Reddit 67% men).
- Joy Buolamwini and Gebru's "Gender Shades" paper: facial recognition misclassified darker-skinned women up to 34.7% vs 0.8% error for white men; Buolamwini founded the Algorithmic Justice League.
- Real-world harms cited: wrongful arrests from facial recognition (Robert Williams, Porcha Woodruff), Amazon's scrapped male-biased recruiting tool, PredPol targeting Black/Latino neighborhoods.
- Rumman Chowdhury led Twitter's META ethics team (found right-wing amplification), calls LLMs "spicy autocorrect," founded Humane Intelligence; critiques "AI Doomer" narrative as self-serving and demographically narrow.
- Argument: present harms (bias, exploited data labelers earning ~$1.32/hr, marginalized-community lockout) matter more than hypothetical extinction risk pushed by figures like Geoffrey Hinton; Gebru founded DAIR.
Connections: Timnit Gebru · Joy Buolamwini · Rumman Chowdhury · Algorithmic Bias · AI Ethics