Researchers introduce ARMove, an AI framework that predicts where people will go next with transparent reasoning instead of black-box calculations

Researchers at an unnamed institution published ARMove, a new system that uses LLMs (AI models trained on text) to forecast human movement patterns — where someone will travel, which route they'll take, or which location they'll visit next. Unlike existing prediction systems that hide their reasoning, ARMove shows its work step-by-step, making it easier to audit whether predictions are trustworthy or biased.

ARMove learns from four types of information at once: general knowledge about how people move, details specific to individual users, regional differences in travel behavior, and new data that arrives over time. The system automatically reweights which information matters most for each prediction, so a forecast for a commuter in Tokyo adapts differently than one for a rural farmer—without requiring separate models for each group.

City planners, transit agencies, and logistics companies can now deploy a single prediction model across different regions without retraining from scratch, saving months of data collection and tuning. Researchers claim the interpretable reasoning also helps compliance teams at ride-sharing or delivery apps explain to regulators why they're predicting a user will be in a certain place at a certain time.