- The user's goals and preferences are not directly observable. They must be inferred from the user's actions.
- The user's goals and preferences may change over time as their situation changes. The assistant must be able to adapt its user model.
- There is uncertainty in inferring the user's goals from limited observations of their actions. The assistant's inferences should reflect this uncertainty.
- For the assistant to be helpful, harmless and honest, its user model and reasoning process must be understandable and adjustable by the user.
To address these requirements, FILUM (Fuzzy Interpretable Learner for User Modeling) uses fuzzy rule-based reasoning over a probabilistic user model. The key assumptions are: