TYPES OF AGENT MEMORY

LEARNING GOALS In this chapter, you will learn the main kinds of memory an agent can have. You will learn the difference between short-term and long-term memory, between episodic and semantic memory, and what it means for an agent to remember across separate sessions. AN EVERYDAY PICTURE OF MEMORY Think about how you remember things. Right now, you are holding in your head the last few sentences you read. That is a small, fast, temporary kind of memory, and it fades quickly. You also remember your own phone number, how to ride a bicycle, and the name of your first teacher. Those are stored long-term, and you can call them up years later. Among your long-term memories, some are stories about particular events, like your last birthday party, and others are plain facts, like Paris being the capital of France. You remember both kinds, but they are different. The same distinctions turn out to be very useful for designing agent memory. One extra point matters for agents: the model itself does not automatically remember anything. All memory in an agent system is something the harness builds, by choosing what to store, where to store it, and what to put back in front of the model.