Explaining Genetic Programming Trees using Large Language Models

LLMs can enhance AI explainability: The paper demonstrates that large language models like ChatGPT can be effectively combined with genetic programming to generate more intuitive, narrative-based explanations of complex AI processes, moving beyond traditional technical documentation to make AI systems more understandable to non-expert users.

Prompt engineering is critical for reliable AI explanations: The research emphasizes that "the importance of prompt engineering in eliciting accurate and pertinent responses from LLMs" is essential—meaning how you ask an AI to explain something significantly affects whether the explanation is accurate and useful, highlighting a key responsibility for those deploying these systems.

LLM hallucinations remain a serious concern: The paper explicitly identifies "hallucinatory outputs" as an important consideration when using LLMs for explanations, warning that these systems can generate plausible-sounding but false information, which is particularly problematic when users rely on AI explanations to understand critical decisions.

Data privacy requires careful attention in explanation systems: The authors flag data privacy as a key consideration when using LLM-powered explanation tools, suggesting that organizations must be cautious about what data they feed into these systems and how explanations might inadvertently expose sensitive information.

User-centered design matters for responsible AI: The introduction of the "user-centred" GP4NLDR dashboard suggests that responsible AI development requires actively designing explanation systems with end-users in mind, rather than assuming technical explanations alone will suffice for accountability and trust.