This is a problem for curious agents that explore by seeing new things. TV shows the agent new things, so they learn a lot by simply looking at the image.
Videos from OpenAI blog (external link).