Geoffrey Hinton fears that we aren't doing enough to ensure the safe development of AI, with military applications posing the biggest threat.