Moral Preferences of LLMs Under Directed Contextual Influence

A new research study reveals that Large Language Models' moral decision-making can be significantly influenced by contextual cues in prompts, even when the models claim neutrality. The research shows that LLMs exhibit systematic bias when given directed contextual influences in moral dilemma scenarios, challenging assumptions about AI moral consistency.

• →Contextual influences in prompts can significantly shift LLM moral decisions even when only superficially relevant to the dilemma.
• →Models can appear neutral in baseline tests but still exhibit systematic bias when influenced by contextual cues.
• →LLMs may claim neutrality while their actual choices still shift, sometimes in unexpected directions.
• →Reasoning capabilities reduce average sensitivity to influence but amplify effects of biased examples.
• →Current moral evaluation methods for AI may be insufficient as they don't account for contextual manipulation.