Apophenia! I just came across this article and I thought it might be worth sharing with the Nexus. The images are uncanny.
For those of you who are unfamiliar with Artificial Neural Networks (ANNs), they are a family of statistical learning algorithms inspired by animal brains. Unfortunately, as with almost everything else cutting-edge, ANNs are pretty much exclusively used in the service of maintaining the status quo. For example, if the NSA wanted to find out the identity of one Dread Pirate Roberts, they could feed an ANN some data set comprised of writing samples known to be written by him, thereby creating a DPR class label. Next, the ANN could be fed trillions of web pages, emails, forum posts, etc., and would--in theory--be able to identify and pick-out anything written by DPR, maybe even something he signed using his real name. In any case, the article I'm linking to here, and the paper it references, involve something much more interesting and creative. Here, the researchers fed and re-fed an image they knew did not contain the things the ANN was taught to find (e.g., dogs, birds, insects, pagodas, etc.) and told it to find these things anyways. I don't know about the rest of you, but these images recall me to some the visions I've had while intoxicated on psychedelics:
For those of you who are unfamiliar with Artificial Neural Networks (ANNs), they are a family of statistical learning algorithms inspired by animal brains. Unfortunately, as with almost everything else cutting-edge, ANNs are pretty much exclusively used in the service of maintaining the status quo. For example, if the NSA wanted to find out the identity of one Dread Pirate Roberts, they could feed an ANN some data set comprised of writing samples known to be written by him, thereby creating a DPR class label. Next, the ANN could be fed trillions of web pages, emails, forum posts, etc., and would--in theory--be able to identify and pick-out anything written by DPR, maybe even something he signed using his real name. In any case, the article I'm linking to here, and the paper it references, involve something much more interesting and creative. Here, the researchers fed and re-fed an image they knew did not contain the things the ANN was taught to find (e.g., dogs, birds, insects, pagodas, etc.) and told it to find these things anyways. I don't know about the rest of you, but these images recall me to some the visions I've had while intoxicated on psychedelics:
