There are two main difficulties in the value alignment problem of AI. The first is the technical difficulty of making machines understand, adopt, and retain human goals. The second difficulty is the philosophical challenge of determining whose values to implement, as there is no global consensus on values.