As we build systems whose capabilities moreand more resemble those of humans, despite thefact that those systems work in ways that arefundamentally different from the way humanswork, it becomes increasingly tempting to an-thropomorphise them. Humans have evolved toco-exist over many millions of years, and humanculture has evolved over thousands of years tofacilitate this co-existence, which ensures a de-gree of mutual understanding. But it is a seriousmistake to unreflectingly apply to AI systems thesame intuitions that we deploy in our dealingswith each other, especially when those systemsare so profoundly different from humans in theirunderlying operation