Part Overview
Prediction and control are components; an intelligent temporal system is the assembly. Part VII steps up a level of abstraction to ask how the pieces from earlier parts combine into systems that reason about time, remember and use tools, and operate where space and time interact. Its three chapters cover the capabilities that separate a model from an agent: causal understanding rather than mere correlation, persistent memory and planning, and the spatio-temporal structure of the physical world.
Chapter 30 develops temporal reasoning and causality: temporal logic and interval reasoning, causal discovery for time series (Granger, PCMCI, causal nets), causal effects of treatments over time, and counterfactual what-if analysis. This is the chapter that lets a system answer not only what will happen but what would happen if. Chapter 31 turns to temporal AI agents: agent architectures, short and long-term memory systems with retrieval, planning and tool use, long-horizon credit assignment, and the hard problem of evaluating agents fairly. Chapter 32 grounds everything in space with spatio-temporal intelligence: mobility and traffic forecasting, video understanding, activity recognition, and dynamic spatio-temporal graph learning.
This part draws threads together rather than starting new ones. The memory systems of Chapter 31 build on the recurrent and belief-state agents of Part VI; the causal discovery of Chapter 30 sharpens the correlational forecasting of Part II; and the dynamic graphs of Chapter 32 extend the event and temporal-graph modeling of Chapter 18. The industrial sensor running dataset returns in the spatio-temporal setting of Chapter 32. Part VII is where the book's components become a coherent intelligent system, setting up the final question of whether such systems can be trusted and deployed.
Temporal logic, causal discovery for time series, treatment effects over time, and counterfactual what-if analysis.
Agent architectures, memory and retrieval, planning and tool use, long-horizon credit assignment, and fair evaluation.
Mobility, traffic, video, activity recognition, and dynamic spatio-temporal graphs where space and time interact.
Where This Part Leads
Intelligent systems are only useful if they can be trusted and run in production. Part VIII: Trustworthy and Deployed Temporal AI builds on the systems assembled here, subjecting them to interpretability, robustness, fairness, and privacy scrutiny, then turning to the serving, monitoring, and MLOps practice that keeps them reliable in the field.