Collaborative World Modeling
Abstract
Major Goals: The current state of the art in world modeling involves such manual effort that it is infeasible to build large models fast enough to enhance short-term decision making about issues of national and global security. We propose an effort to automate large parts of the building and running of such models, enabling an order of magnitude reduction in development time and a significant increase in the scale of problems that can be explored. The key insight is that developing large-scale world models requires a collaboration between human analysts(providing intuition, strategic thinking and thinking out of the box) and automated intelligent systems (providing large scale data search and analysis, planning and running simulations). By harnessing human-machine collaboration, we enable rapid development of models that go far beyond what humans or automated methods could each achieve alone. The key to success is that model development is iterative, with extensive interaction between the analyst and the system throughout. We will leverage our existing technology base of (1) collaborative problem solving through human-machine interaction, (2) planning and (3) task learning, to build an open, extensible and interactive system for World Modeling. Quantitative Reasoning Engines (QREs) can be plugged in as they become available (or taken offline), using a uniform messaging structure for declaring their inputs, outputs and data requirements. The parameters will be encoded in a standardized ontology, and matched with natural language text that might describe these parameters using a module that scores the semantic similarity between the parameter documentation and the natural language text. The system will use this information, together with information obtained from reading systems or from interaction with Subject Matter Experts, to build and continually update its causal models.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 12, 2022
- Accession Number
- AD1221386
Entities
People
- James F. Allen
Organizations
- Florida Institute for Human and Machine Cognition