The Word AI Couldn't See in Google Photos
In 2015, Google's image recognition system labeled photos of Black people with a racist slur, and the company's immediate fix was to delete the category entirely. This episode traces how a single mislabel exposed the gap between machine learning promises and training data reality, and why erasing a symptom is easier than solving the underlying bias.
Listen in the Fylom app.
Google Photos mislabeled Black individuals as gorillas due to machine learning failures with darker skin tones.
Supervised learning models mirror human prejudice when training datasets lack diverse representation of different groups.
Google suppressed the issue by blocking primate-related search terms instead of retraining the underlying model.
Technical workarounds like deleting labels failed to fix the structural biases within the neural network.
Professional tools like the Cloud Vision API retained problematic labels long after the public app fix.
One-fifth of training images contain multiple objects, forcing models to make flawed single-label guesses.
- 01Intro1 min
- 02The Notification2 min
- 03The Mirror of the Dataset3 min
- 04The Deletion Fix2 min
- 05Symptoms vs. Systems3 min
- 06Outro1 min
- Google apologises for Photos app's racist blunder
- Google says sorry for racist auto-tag in photo app
- Incident 16: Images of Black People Labeled as Gorillas
- Report 95 - AI Incident Database
- Google Photos called Black woman a ‘gorilla': AI racism row reignites as user says 'not racist, just stupid'
- Google 'fixed' its racist algorithm by removing gorillas from ...
- Google Apologizes After Photos App Autotags Black People as ...
- Google Tags Black People As "Gorillas"
- Google's Photo App Tags Photos Of Black People As 'Gorillas' (2015)
- [PDF] From ImageNet to Image Classification: Contextualizing Progress on ...
- Understanding and Mitigating Annotation Bias
- arXiv:2107.07919v1 [cs.CV] 16 Jul 2021
- Automated Classification of Model Errors on ImageNet
- Google Photos Labels Black Users as Gorillas — Fix: Remove Gorillas from Labels
- Google apologizes after its Vision AI produced racist results - AlgorithmWatch
Fylom generates episodes like this on any topic you're curious about.
Fylom episodes are researched, written, and voiced by AI. Automated checks help catch inaccuracies, but episodes aren't reviewed by a human and AI can still get things wrong. Treat them as a starting point, not a source of record — more in our accuracy disclaimer.