Process Knowledge-infused Learning for Suicidality Assessment on Social Media

Process knowledge improves human understanding of AI decisions: The paper demonstrates that when AI systems explain their predictions using the actual clinical or expert processes used to make decisions (rather than just analyzing the data), humans understand these explanations better, achieving an annotator agreement score of 0.72.

Traditional AI explanations fall short for real-world users: Current explainable AI methods that highlight important words or features in text may satisfy computer scientists but don't adequately communicate reasoning to end-users like healthcare professionals who need to understand why a decision was made in terms they recognize from their own work.

Process-based AI is competitive with standard approaches: The paper shows that incorporating structured process knowledge doesn't sacrifice performance—the new Process Knowledge-infused Learning (PK-iL) method performs competitively with state-of-the-art baseline models while providing better explanations.

Responsible AI requires bridging the explanation gap: For AI adoption in high-stakes domains like healthcare and mental health assessment, systems must explain decisions using human-understandable concepts aligned with how experts actually work, not just through data-driven pattern analysis.

Application to suicide risk assessment demonstrates clinical relevance: The research applies this approach to detecting suicidality indicators on social media, a domain where transparent, trustworthy AI could meaningfully assist healthcare professionals at scale while maintaining accountability.