GPT versus Humans: Uncovering Ethical Concerns in Conversational Generative AI-empowered Multi-Robot Systems

GPT systems may miss emerging ethical concerns: When asked to deliberate on ethical issues in multi-robot systems, GPT agents primarily reproduced existing AI ethics guidelines rather than identifying novel problems. In contrast, human experts raised new concerns about "deviance, data privacy, bias and unethical corporate conduct" that weren't captured by the AI system, suggesting LLMs may have blind spots regarding context-specific ethical challenges.

Significant gap exists between AI and human ethical reasoning: The study directly compared GPT agents with 16 human experts across workshops and found meaningful differences in how each group identified and prioritized ethical concerns. This gap indicates that relying solely on AI systems to identify ethical problems in complex applications like multi-robot systems could leave important issues unaddressed.

LLMs tend to recycle rather than innovate on ethics: GPT agents emphasized concerns "present in existing AI ethics guidelines" rather than generating original ethical insights. This suggests that while LLMs can articulate established ethical principles, they may struggle to anticipate novel risks that emerge from new technological combinations—a critical limitation for responsible AI development.

Multi-robot systems powered by LLMs create compounded ethical challenges: The paper highlights that applying LLMs to already-complex systems like multi-robot coordination introduces "novel ethical challenges" beyond those of LLMs alone. This suggests that ethical review processes need to account for how AI systems interact with other technologies, not just evaluate them in isolation.