The nightmare of robots controlling the human race will not come true, says Richard Waters. The future of AI will see semi-autonomous systems rely on close cooperation with people, uniting machine learning and human judgement. But there are dangers in robots leading humans astray