Lesson 1 — Activation and decay

Duration: ~45 min · Mode: shared read-only project (classroom) · Instrument: the web UI Graph and Query tabs

Learning objectives

By the end, students can:

1. Explain activation as a use-dependent strength score on a memory.
2. Describe exponential decay and connect it to human forgetting curves.
3. Predict which memories a system (or a person) will recall fastest, and why.

The science in one paragraph

Human memory retrieval is use-dependent: the more often and more recently something was needed, the faster it comes back. Ebbinghaus measured the forgetting curve in 1885; a century later, Anderson & Schooler (1991) showed retrieval odds track the statistics of the environment, and ACT-R (Anderson et al., 2004) formalized it: each memory chunk carries an activation value that grows with use and decays with time. Memory Layer implements exactly this — every query result boosts a memory's activation (citations in an answer boost more), and between uses the score decays exponentially with a 30-day half-life. The self-maintaining memory page documents the mechanism.

Prep (2 min)

Stack up, classroom seeded, student mode on. Projector on the Graph tab with the Code and Provenance layers toggled off, Memory layer on.

Guided activity

1. Read the instrument (5 min). Node size and warmth encode decayed activation. Freshly seeded, the corpus is nearly uniform — this is the "before" picture. Ask students to nominate which memory should become hottest if the class spends an hour asking about reinforcement.

2. Heat something up (10 min). Every student, on their own machine, opens the Query tab and asks three questions of their choice about how the system remembers, e.g.:

• "How does reinforcement work?"
• "What happens to memories that are never used?"
• "What does a good memory look like?"

Each answer cites its evidence. Point out that every citation just boosted that memory's activation — the class is collectively rehearsing.

3. Observe (5 min). Refresh the Graph tab on the projector. The cited memories are now visibly larger and warmer than the rest. Click one and read its activation value; compare with an unqueried memory near zero.

4. Decay thought experiment (10 min). The system's half-life is 30 days: an untouched memory keeps half its activation after a month, a quarter after two. Sketch the curve on the board and have students plot where "yesterday's password" and "your childhood phone number" sit on their own curves — rehearsal count vs. recency.

5. Why forget at all? (10 min discussion). Storage is cheap; the system could keep everything equally strong. Ask why it deliberately lets things fade. Target insight: retrieval is a ranking problem — forgetting is how both brains and this system keep the relevant things on top (Anderson & Schooler's "rational analysis" argument). Note the saturation rule: activation breaks ties but is capped so popularity can never beat relevance.

Exit ticket

One sentence: "Name something you remember effortlessly and explain its activation in terms of use and recency."

References

• Ebbinghaus, H. (1885). Über das Gedächtnis — the original forgetting curves.
• Anderson, J. R. & Schooler, L. J. (1991). Reflections of the environment in memory. Psychological Science, 2(6).
• Anderson, J. R. et al. (2004). An integrated theory of the mind. Psychological Review, 111(4) — ACT-R.