Building Trustworthy AI: Transparent AI Systems via Large Language Models, Ontologies, and Logical Reasoning (TranspNet)

LLMs alone aren't transparent enough for high-stakes decisions: The paper identifies that while Large Language Models generate accurate outputs, their "black box" nature creates significant barriers to transparency and trust in critical fields like healthcare and finance where explainability is essential.

Combining symbolic AI with LLMs improves trustworthiness: TranspNet integrates domain expert knowledge, retrieval-augmented generation (RAG), and formal reasoning frameworks like Answer Set Programming (ASP) with LLMs to add structured reasoning and verification capabilities that make AI decisions more interpretable and verifiable.

Structured reasoning enables accountability: By incorporating formal reasoning frameworks, the system can provide step-by-step logical justifications for its outputs, helping organizations meet regulatory expectations for transparency and accountability in AI decision-making.

Expert knowledge is crucial for reliability: The approach leverages domain expert knowledge as a foundational component, suggesting that trustworthy AI systems require human expertise to be embedded into their design rather than relying solely on data-driven models.

Hybrid approaches address real-world trust requirements: TranspNet demonstrates that combining multiple AI techniques—rather than relying on a single approach—creates systems that are both accurate and interpretable, making them suitable for real-world applications where stakeholders need to understand and trust AI decisions.