But how do we guarantee that they behave? This is called the problem of alignment, making sure AI does what we tell it to with no …

source