Something just happened in the room. A hand went up too fast, a table went quiet, or a learner laughed at the wrong moment. If you can name what that was, out loud, and route it back into the lesson, you have turned a prediction error into a model update. If you cannot, the same signal reads as disruption. This post is a working vocabulary for that naming. It sits inside the cluster opened in the primer, how prediction shapes experience.
Three flavors worth naming
In the conceptual language of active inference, the learner is always running a forecast of what the next moment will feel like, and the gap between forecast and outcome is prediction error. Class E The technical grounding is Parr, Pezzulo and Friston, Active Inference (MIT Press, 2022). What matters in a room is that prediction error does not arrive labeled. It arrives as feeling. Three flavors show up most often.
- Surprise. The world did not do what the learner expected. Eyes widen, a hand moves before a mouth does, someone says "wait, really?" The mismatch is on the surface and the learner is oriented toward it.
- Confusion. The mismatch is present but the direction is unclear. The learner cannot yet tell whether their model is wrong or the material is wrong. Faces tighten, notes slow, the room gets quiet in a specific way that is not the quiet of thinking.
- Delight. The world did something better than expected, and the learner's model just got a positive update. A grin, a small exhale, a "huh." Delight is prediction error too, and it is the easiest one to squander by moving on too fast.
Each is data. Each tells you the update rate is high right now, which is the window in which teaching actually lands. The move is to name the signal, out loud, in words the group can use, and decide together what to do with it.
Four sentences that work with a group
You do not need clinical language. You need short, specific sentences that route the signal back into shared attention.
- "Something just surprised somebody. Whose model just got a hit?" Gives the learner cover to own the update without confessing weakness.
- "There is a confusion in the room. Let us find its shape before we solve it." Slows the reflex to answer and turns confusion into a shared object.
- "That was delight. What did you just see?" Delight evaporates. Naming it makes the pattern explicit, so more than one person keeps it.
- "I predicted this section would land differently. It did not. That is data for me, not for you." Model the move; adults route prediction error the same way you are asking learners to.
Two moves that collapse the signal
- Explaining the surprise away. When a learner is genuinely surprised, the temptation is to smooth it. You just answered a question they had not asked yet. Let the surprise sit for a beat. Ask what they think it means.
- Rescuing the confusion. A rescued confusion is one that never got mapped, which means the underlying model did not update, which means the same confusion returns next week. Name it, hold it briefly as a shared object, and let the group articulate what specifically is unclear before you resolve it. Class C The same pattern is built into the workshop, where the room's confusions are staged as artifacts we can point at.
What this is not
Two fences. Naming prediction error is not clinical care, and it is not a substitute for the support a clinician provides when a nervous system needs one. It is a pedagogical practice, trauma informed and non-clinical. And it is not easy on the first try. Rooms get better at it the way any group gets better at a shared language, by using it and correcting each other.
The one page version
When the temperature of the room changes, ask which flavor of prediction error just showed up, and name it in plain words the group can use. Surprise, confusion, or delight. Hand the signal back to the room as shared attention, not as a problem to be managed away. You will be teaching the room to run its own update loop on purpose, a skill that outlasts any single lesson.
How prediction shapes experience
The three sentence primer this post sits inside. Start here if prediction error is a new phrase.
Next in the clusterThe safety first update rate
Why felt safety is the condition under which naming prediction error actually lands, and how to build it into a lesson plan.
The workshopPractice these moves with us
The two day workshop where teams rehearse naming prediction error in their own rooms, with support, until the practice sticks.
Evidence classes used in this post: Class E expert citation, Class C configuration and integration. Full method and taxonomy at the transparency page.