A study found that people were routinely unable to accurately distinguish between AI and human generated poetry - getting it wrong more often than not