Interleaving, Dual Coding & Desirable Difficulties

Here is one of the strangest truths in all of learning science: the study methods that feel the best are usually the ones that work the worst, and the methods that feel slow, clumsy, and frustrating are often the ones that build memory that lasts. This chapter is about three of those "feels-bad-but-works" ideas, and about why your own sense of "I've got this" is one of the least trustworthy signals you have.

If you are building an artificial-intelligence tutor (a computer program that teaches one learner at a time and adapts to them), this chapter is also a warning label. Learners will resist the methods that help them most and may even rate your tutor poorly for using them. A good tutor must use these methods anyway — and explain why.

8.1 Interleaving: mix it up instead of drilling one thing

Let's define two ways to practise.

Blocked practice means you drill one type of problem over and over before moving on to the next type. If we label problem types A, B, and C, blocked practice looks like AAAA BBBB CCCC. You do all the addition, then all the subtraction, then all the multiplication.

Interleaved practice means you mix the types together so you never know which is coming next: A B C B A C…

Blocked practice feels wonderful. You get into a rhythm, your answers come faster and faster, and you finish feeling like a genius. Interleaving feels terrible by comparison — you stumble, you mix things up, you feel slower. And yet, in a famous 2010 study by Kelli Taylor and Doug Rohrer, students who practised maths with interleaving scored roughly double on a later test compared with students who used blocked practice.

Why such a huge gap? Because of a skill that blocked practice never lets you build: choosing the right approach. When all the problems on a page are the same type, you already know the method before you even read the question. You are practising the execution of a method, not the decision of which method to use. But real tests and real life never tell you the type in advance. Interleaving forces you to look at each problem fresh and ask "what kind of problem is this?" — which is the actual hard part.

There is a free bonus, too. Because interleaving spreads your practice of each type across the whole session instead of cramming it into one block, you automatically get spacing (the gaps-between-practice effect from the previous chapter) layered on top. Two good things for the price of one.

One caution: interleaving is for after the basics are in place. You first teach what type A even is; then you start mixing it with B and C. Mixing types a learner has never seen just creates confusion, not productive challenge.

8.2 Dual coding: words plus the right picture

Psychologist Alan Paivio proposed that the mind handles information through two separate but connected channels: a verbal channel (words you read or hear) and a visual channel (images and diagrams). "Dual coding" simply means using both at once.

Because the two channels are separate, presenting an idea as both clear words and a matching picture gives your brain two routes to store the memory and two routes to find it again later. It also spreads the mental effort across two channels instead of jamming everything into one. Richard Mayer turned this into the Cognitive Theory of Multimedia Learning, and the headline finding is steady: people learn better from words plus relevant pictures than from words alone.

But there is a giant catch, and it is the part most people get wrong.

8.3 Desirable difficulties: the umbrella idea

Robert Bjork coined the term desirable difficulties in 1994 for learning conditions that make studying feel harder and slower right now but produce stronger, longer-lasting learning later. This is the big idea that ties this whole part of the book together, because the powerful techniques you've met are all desirable difficulties:

• Retrieval practice — pulling an answer out of your head instead of rereading it.
• Spacing — spreading practice out over time instead of cramming.
• Interleaving — mixing problem types instead of blocking them.
• Generation — trying to produce an answer before you're told it.
• Variation — practising in changing conditions rather than identical ones.

Every one of them slows you down and makes you stumble in the moment. And every one of them builds memory that endures and transfers to new situations.

Now the single most important word in the phrase: desirable. A difficulty only helps if the learner has enough background knowledge to overcome it through effort. Strip away the prerequisites and the very same difficulty becomes undesirable — pure, pointless frustration that teaches nothing.

This is exactly where an AI tutor can shine or fail. It must deliberately introduce productive struggle — ask before telling, mix the problems, space the reviews — while constantly checking the learner's level so the difficulty stays in the helpful zone and never tips into "I have no idea what's going on and I want to quit."

8.4 The illusion of fluency: why easy lessons are a trap

Here is the human-factors problem that makes all of this hard to sell. We judge how well we've learned something by how easy it feels to process. And ease is a liar.

Rereading a chapter and running a highlighter over it make the text feel smooth and familiar. That smoothness creates a strong, convincing feeling of "I know this." But that feeling is mostly recognition of the surface — "I've seen these words before" — not real, retrievable knowledge you could produce on a blank page. Surveys find that about 80% of college students name rereading as their top study method, even though it is one of the least effective things you can do. Cramming is the same trap: it feels productive, it might even pass tomorrow's quiz, and then the material evaporates within days.

So we get a cruel mismatch: the effective methods feel worse while you're doing them, so learners conclude they're working less well and abandon them for the comfortable, useless ones.

   FEELS EASY  ----but---->  LEARNS LITTLE
   (rereading, highlighting, cramming, blocked drills)

   FEELS HARD  ----but---->  LEARNS DEEPLY
   (retrieval, spacing, interleaving, generation)

   The feeling and the result point OPPOSITE ways.

8.5 Putting it together for an AI tutor

The table below sums up the "easy trap" versus the "harder path" — and what a well-built tutor should actually do.

Because the good methods feel worse, your tutor can't just use them silently — it must also counter the fluency illusion. Briefly explain to the learner why the harder path works, and show them objective evidence of progress (their real retrieval success over time), so their confidence is anchored to what they can actually recall, not to how smooth a lesson felt. And remember the deepest rule of all: judge whether learning worked on a delay, not in the moment. The technique that looks weakest during practice is usually the one that wins the test two weeks later.