Google researchers achieved nearly 91% smile-detection accuracy on the Faces of the World (FotW) benchmark by training racial and gender classifiers as part of their model.
Sources
- Google’s New AI Smile Detector Shows How Embracing Race and Gender Can Reduce Bias (technologyreview.com)
- https://ai.google/research/pubs/pub46513/ (ai.google)
- https://www.theregister.com/software/2017/12/06/google-learns-to-smile-because-ais-bad-at-it/798222 (theregister.com)
- https://www.techmonitor.ai/digital-economy/ai-and-automation/google-ai-diversity (techmonitor.ai)
- https://arxiv.org/pdf/1712.00193 (arxiv.org)
More in Technology
- Accelerating change is a concept in futures studies and the history of technology describing the observed exponential nature of the rate of technological change in recent history.
- Without instruction fine-tuning on biomedical data, Llama3-RankRAG performed comparably to GPT-4 on five biomedical RAG benchmarks.
- AI melanoma-diagnosis systems perform better on white skin than black skin, largely because their training data contains more examples of white skin.
- FrontierMath is a math benchmark whose difficult questions remain unpublished so that AI companies cannot train their models against them.
- Apollo lunar-module astronauts stood side by side only 16 inches apart, viewing the lunar surface through two two-square-foot windows.
- Google Gemini generated historically inaccurate images depicting racially diverse US founding fathers and Black Nazis, despite using bias-reduction techniques that could score well on existing fairness benchmarks.