A Human-centred Future: Designing Safe AI Systems