Saying that AI will not learn how to cope with the social aspects of driving on the basis of current driverless cars is a bit like saying planes will never cross the Atlantic on the basis of the early flights by Orville and Wilbur Wright.
We are at early days and the progress is phenomenally fast. Experts such as Geoffrey Hinton are saying the latest AI models are doing things they didn’t expect to see for decades. My own experience of ChatGPT is that it is both astonishingly brilliant and fantastically stupid at the same time. It’s been likened to the best read 10 year old there has ever been who still doesn’t understand basic aspects of the real world. But for how much longer? Nobody is saying ChatGPT will change the world, or endanger it, it’s the trajectory of the progress which is the amazing and/or alarming thing. And GPT 4 is already here.
AI is now better at diagnosing patients than human doctors. And a recent study shows AI interactions are perceived as more empathetic than human doctors too.