Analogies, Diagrams, Animations & Simulations

So far we have talked a lot about what a tutor teaches and when it reviews. This chapter is about how an idea gets into a learner's head in the first place — through the words, pictures, and moving things we put in front of them. A great explanation is not just "correct"; it lands. It connects something brand new to something the learner already owns. That is the job of analogies, diagrams, animations, and simulations.

Before we start, one plain-English definition we will lean on the whole chapter. Working memory is the tiny mental "desk" where you do all your active thinking. It is shockingly small — careful research (Nelson Cowan, 2001) puts it at about four chunks of new information at once, and it empties in seconds if you look away. Every teaching aid in this chapter either helps fit an idea onto that small desk, or accidentally clutters it. That single idea decides whether a fancy animation is a gift or a distraction.

14.1 Why analogies work

An analogy says "this new thing works like that familiar thing." It works because of how memory is built. Long-term memory stores organized patterns called schemas — think of them as ready-made mental folders ("how a bank account works", "how a bucket leaks"). When you say "an Ease Factor is like the gear ratio on a bike", you are not making the learner build a brand-new folder from scratch. You are letting them borrow a folder they already have and bolt the new idea onto it. The new material arrives as one big familiar chunk instead of five strange little ones, so it slides through that narrow four-chunk doorway.

This is also why analogies feel "clicky." The effort of understanding (the good kind, called germane load — the mental work of actually building understanding) goes toward mapping, not toward decoding strange new pieces one at a time.

14.2 How analogies fail

Every analogy is a partial match. It lines up in some ways and breaks in others — and the learner usually cannot see where the break is. Taken too far, the borrowed folder smuggles in wrong ideas.

Three rules keep analogies honest for an AI tutor:

• Name the limit. A great explanation says where the analogy stops: "Working memory is like a computer's short-term memory — but unlike a computer it also wipes itself clean in seconds." Stating the edge prevents the wrong import.
• Match the learner's world, not yours. An analogy only works if the "familiar" island is genuinely familiar. A cricket analogy lands for one learner and confuses another. This is where a tutor that knows the learner can tune the comparison to their interests — a real advantage over a fixed textbook.
• Retire it as expertise grows. Beginners need the bridge; experts find it clutter. The same pattern shows up everywhere in learning: support that helps a novice can slow down someone who has moved past it.

14.3 Diagrams and dual coding

Dual coding (Alan Paivio; turned into practical advice by Richard Mayer) is the finding that the mind has two separate channels for taking information in: a verbal channel for words, and a visual channel for pictures. Because they are separate, a well-matched word-plus-picture pair gives the brain two routes to store and later find the same idea — and spreads the work across two desks instead of overloading one. People reliably learn more from relevant words plus relevant pictures than from words alone.

But dual coding has a giant catch, and it is the most common way teaching media backfires: it only works when the picture carries real information and sits right next to the words it explains. Violations actively hurt:

Here is the kind of small, information-carrying diagram a tutor should generate — every box and arrow means something:

  NEW IDEA  ->  [ Working Memory ]  ->  LONG-TERM MEMORY
  (the words)     desk: ~4 chunks       (vast, durable)
                       ^
                       |
              ANALOGY + DIAGRAM
        (pack ideas into fewer, bigger chunks
         so more fits through the doorway)

14.4 When animation or simulation beats static text

An animation shows change over time. A simulation is an animation the learner can poke — change an input and watch the result. They are powerful but expensive, so use them only where a still picture genuinely cannot do the job. Reach for motion or interaction when:

• The thing to learn IS the change over time. How a wave travels, how a sorting algorithm rearranges a list, how blood circulates — these are processes, and a frozen frame hides the very point.
• Cause and effect needs to be felt. A simulation where the learner drags the "desired retention" slider and watches review intervals shrink teaches the trade-off in their bones, in a way a paragraph cannot.
• The space is too large to describe. Exploring "what happens at every interest rate" is better roamed than read.

14.5 The cost/benefit: build rich media, or generate it?

For an AI learning platform this is a real budget decision, not a detail. You have three rough options for any visual:

The practical pattern that falls out of this: generate the words, verify the pictures. Today's language models are excellent at producing analogies and prose explanations and adapting them to a learner's level — that is the cheap, high-value win. They are noticeably worse at diagrams: they struggle with geometry, proportion, and spatial layout unless the task is first turned into structured text. So treat every generated diagram or math figure like a junior author's first draft — render it, look at it, and run it past a check (or a human) before it reaches a learner. Reserve expensive hand-built simulations for the few "hero" concepts where exploration is the whole point and a static image truly fails.