XAI (Explainable AI)
Investigators, analysts and cybersecurity teams reviewing AI-generated investigative information
Category
Explainable AI
Owner partner
HUA
Related use cases
UC1.1, UC1.2
Tool summary
What does an AI-generated finding mean, and what information supports it? GANNDALF’s XAI component presents explanations in accessible text or visuals, helping investigators understand and assess results without specialist AI knowledge.
The challenge
AI tools can produce findings that are difficult for investigators without specialist AI knowledge to interpret. When the reasoning or supporting information is unclear, analysts may struggle to assess a finding or communicate it to colleagues. Clear textual and visual explanations can help them understand the output and use it appropriately during an investigation.
What it does
The XAI component takes structured knowledge triples generated by KModelC and produces short explanations grounded in the source report. Each explanation remains linked to its corresponding triple, allowing analysts to examine why a fact appears. The pipeline also produces explanation embeddings for later processing. Then, visual counterfactuals are developed with synthetic images to illustrate specified changes to people or objects while preserving the surrounding scene.