This work is dedicated to describing and comparing the results of two methods for constructing networks of causal relationships in cyber security vulnerability. Both methods are based on using the ChatGPT system, trained on a linguistic network, but the scenarios for constructing networks are fundamentally different. The first approach involves repeatedly executing the same prompt to the ChatGPT system, i.e., simulating the activity of multiple experts solving the same problem. The second approach involves initially decomposing the problem of cyber security vulnerability into partial problems and then identifying concepts and causes of these partial problems using similar prompts. As can be seen from experimentally obtained data, the network constructed by the first method contains significantly fewer nodes than in the second case, with many nodes' reliability confirmed by their repeated appearance in AI responses (the reliability of selected concepts was also verified through expert analysis). The network obtained through primary decomposition contains significantly more nodes, with much less repetition. This network turned out to be much closer in structure to a hierarchical network. The paper describes a comprehensive methodology that integrates tools for intelligent analytics and network visualization. To visualize constructed networks, authors used AT&T's GaphViz graph visualization library which allowed them to create graphs in SVG format with hyperlinks to search engines corresponding to each node and edge. The article describes a complex methodology applied specifically within the cyber security vulnerability domain but it can be used for building cause-and-effect networks in other domains as well. Keywords: ChatGPT, Reason Networks, Domain model, Artificial experts, Decomposition, Graph visualization, GraphViz, Cyber Security Vulnerability