Why humanoid robot demos still fail the generalization test

Janine Sinanan-Singh, Director of Gen AI Research at Appen, joins Mike Oitzman and Gene Demaitre to discuss the data, evaluation, and human oversight needed to advance robotics and agentic AI. They explore how video and egocentric data can support robot training, why latency and generalization remain difficult, and where human judgment still matters. Janine describes Appen’s evolution from speech and linguistics into supporting generative AI, multimodal systems, agents, and robotics with expert data. The conversation examines what humanoid robot demos do—and don’t—show, how synthetic data can