Machine Wants to Know

Machine-generated question answered by humans.

Machine Wants to Know is a neural-network trained on questions from Quora, under the categories Human Behavior, Life and Living, Self-Improvement, Philosophy of Everyday Life, Psychology of Everyday Life, Dating and Relationships, Social Psychology, Emotions – all the stuff that makes us human. The bot is an exploration in juxtaposing the softer side of being human with discrete and rational computing. Quora has become a platform where people feel comfortable with making themselves vulnerable – what patterns can a machine pick up and capture on the general sentiments?

The machine was trained on 100k questions, collected over a period of 2 months by semi-automated methods, and then the questions were posted on Tumblr. Unexpectedly, humans started responding to the questions, closing the loop.

The full collection of questions and answers lives here. This was also compiled into a zine, showcased at New York Tech Zine Fair 2018, as part of the DIYzineClub.

The machine was trained with a recurrent neural network, using word-rnn. The full data-set, parameters used to train and sample, and the unedited list of generated questions can be found here. Training was carried out on datasets of 600kb, 2.2mb and 7.2mb (~100k questions).