"I spent three years building ImageNet when everyone said datasets did not matter — only algorithms did. But the algorithm learns from the data. The data learns from the human who labeled it. The human brings context that no algorithm can generate alone. ImageNet was not a dataset. It was a coordination event between human perception and machine learning."
Computer scientist who catalyzed the deep learning revolution not by building a better algorithm but by creating the right coordination substrate — ImageNet, a dataset of 14 million labeled images that gave neural networks the context they needed to learn visual perception. As co-director of Stanford's Human-Centered AI Institute, she now argues that AI's missing ingredient is the human coordination layer: ethics, embodied context, and social embeddedness that data alone cannot encode.