June 29th, 2015
Episode 151 of 457 episodes
In an effort to create a more accurate image recognition system, researchers have been training Google's neural network to pick out objects obscured by denser and denser layers of optical noise. But by feeding the AI the other direction - giving it nothing but noise and asking it to pick out images - they've discovered a surreal and dreamlike quality in the network that is both beautiful and haunting. Anthony and Jeff marvel at the ability for an AI to "dream", and project how this will lead to our ultimate downfall.