> Since we first started carving notches in sticks, we have used things in the world to help us to know that world. But never before have we relied on things th

> Since we first started carving notches in sticks, we have used things in the world to help us to know that world. But never before have we relied on things that did not mirror human patterns of reasoning — we knew what each notch represented — and that we could not later check to see how our non-sentient partners in knowing came up with those answers This is alarmist and completely untrue. The scientific method specifically involves forming a hypothesis, trial, and error. We don't know much until we experiment. That deep learning makes things possible before we fully understand them is nothing new. I can understand this fearful attitude coming from non-scientists who fear the black-box nature of machines. However, technologists should understand that the functionality of this tech can be reverse engineered, or verified, with additional legwork. We should proceed cautiously and with vigilance. Saying we'll "never" understand the inner-workings of such algorithms is too much.