Users praised the study finding that LLMs increase blame attribution by shaping moral alternatives, calling the work great and looking forward to discussing it in Rio.
We find that LLMs sample option spaces differently from humans, producing higher-valued but more semantically concentrated alternatives. 💭
When people consider action alternatives generated by an LLM, they assign more blame than when they consider human-generated alternatives.
Feat. @neele_engelmann, @amnussberger, Arne Maas and @phillipsjs. Come catch us at @cogsci_soc in Rio🌴 presenting this.
@Lara_Kirfel @neele_engelmann @amnussberger @phillipsjs @cogsci_soc Looks great - see you in Rio to discuss :-)