Bias in AI Models
Moustapha discusses the complexities of mutual information in AI and highlights the troubling biases present in datasets, particularly concerning racial associations. He envisions a future where models can self-critique and iteratively improve to reduce biases, emphasizing that while complete elimination of bias is impossible, significant mitigation is achievable. The conversation also explores the mechanisms that could guide models in focusing on relevant features during training.In this clip
From this podcast
The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)
Security and Safety in AI: Adversarial Examples, Bias and Trust with Moustapha Cissé - #108
Related Questions