Global early warning systems from mass movements have been increasingly important as the triggering rainfall patterns continue to intensify due to climate change. In particular, the Norwegian early warning systems for landslides and avalanches currently use a highly sensitive classification of danger reports that are presented at the county or village level of information, without any fine details needed to understand its potential implications on critical infrastructures such as transportation and communication systems. We developed a novel inter-graph framework that applies artificial intelligence to the interaction of various forms of graphical or relational connections such as (1) the spatial connectivity between road networks and settlements and (2) the proximity and similarity of mapped locations with different hydrological and geological variables, thereby producing near-real-time country-wide susceptibility maps and assessment of exposure level of all settlements in Norway. This study offers an alternative methodology to evaluate the widely-used disaster risk equation by explicitly modelling the interaction of these graphical or relational connections not only between exposure and susceptibility, but also between hazard, exposure, and vulnerability, for a holistic assessment of risk using geospatial datasets and artificial intelligence.