

I think this is bullshit, principally when he says he “mapped” the neural response to actions in the game. You can take any sufficiently large, random, neural network and train the mapping for it to perform a task under reinforcement learning. This is just a reservoir neural net with extra steps




Another thing, Google usually doesn’t “do” anything, at best they sponsor projects. As far as I can tell from the article, the mapping was most likely done by Janelia researchers.