There Is No Universal Best Interface

The previous chapter ended with a temptation. Priya’s behaviour is leaving traces. She opens tables more often than prose, expands the mathematical detail, ignores diagrams. A system that remembers could draw the obvious conclusion: this person is a certain kind of reader. Diagram people get diagrams, detail people get detail, and the interface learns her type and stops asking.

This chapter takes that shortcut apart. It does not do so by saying that people are all the same, which they plainly are not. The target is exact:

A person possesses a reasonably stable representational type, and matching a representation to that type reliably improves how well they do.

Everything else survives the attack: preferences, which are real and deserve respect; strategies people choose for tasks; the properties of materials; the effects of expertise; and accessibility needs, which are another thing entirely. Only the type does not.

What the evidence does and does not say

The standard that any claim about types has to meet was set out clearly in 2008 by Pashler, McDaniel, Rohrer and Bjork, in a review commissioned to assess the learning-styles evidence for education. Liking a form is not benefit from it. The pattern that would justify matching is a crossover: people of one type do better with one representation, people of the other type do better with the other.

TYPE A:  representation A  >  representation B
TYPE B:  representation B  >  representation A

Each group has to do better with its own matched form, not merely like it, and not merely differ from the other on average. The reviewers found no adequate evidence base for deploying learning-style assessments, and that criterion has governed serious work since. A 2024 meta-analysis found a small overall benefit of matching that mostly disappears under the crossover requirement, with low study quality, and its authors themselves recommended against adoption. A 2025 review of work on matching argues that much of the apparent support comes from research that blurs together styles, preferences and strategies. The summary the chapter relies on is deliberately modest: the evidence does not justify converting a stated or inferred preference into a stable representational type and assuming that instruction matched to that type produces better outcomes. It does not say individual differences do not exist. Exact figures are left out here because the point does not depend on them.

Things that are not the same thing

The word “style” is doing the work of several quite different ideas, and the confusion is exactly what lets a type claim look better supported than it is. Separating them gives the chapter its working vocabulary:

STYLE          "I am a visual learner."
               a stable classification of the person; the claim attacked

PREFERENCE     "I like diagrams."
               a legitimate choice, honoured as a choice

STRATEGY       "For this problem I draw it out."
               behaviour selected for a task

AFFORDANCE     "This is easier to see as a graph."
               a property of a representation and a material, true for anyone

EXPERTISE      "This explanation helps a newcomer and is redundant for an expert."
               a relation among a person's knowledge, a domain and a treatment

ACCESSIBILITY  "I need captions, or output that works with a screen reader."
               a requirement, not a style

Each of these is a different object, and the separations the chapter needs follow directly:

preference  ≠  performance
strategy  ≠  trait
expertise  ≠  identity
accessibility requirement  ≠  preference
current condition  ≠  stable type

Most of what looks like fit between a person and a form can be explained without any type at all. Strategy and affordance explain a great deal: people reach for a diagram when the material has topology, and a table when it has columns. Preference explains much of the rest, and preference is something a person can simply state. Where a genuine difference between people does show up, it usually turns out to be about knowledge, and that deserves a closer look.

Expertise changes what helps, and it is scoped

There is one well-supported place where different people really do benefit from different treatments. Research on what is called the expertise-reversal effect finds that learners with little prior knowledge tend to benefit from more instructional support, while learners with a lot of it tend to do better with less. What helps a beginner can become redundant, even obstructive, for an expert.

A beginner may need terminology expanded, intermediate steps shown and relations made explicit. An expert may be better served by compressed notation, the exceptions, the unusual cases and direct access to the evidence. The effect is asymmetric, since supporting novices matters more than unburdening experts, and it is well enough supported to guide design with the usual caveats. No visual or auditory identity appears anywhere in it. The conditioning variable is knowledge, not type.

And even this does not license an “expert type”. Priya is an expert in distributed consensus and a novice in database internals, and the representation that suits her differs between the two on the same afternoon. Expertise is scoped to a domain, and it changes as a person learns, so the treatment that helped her last year is redundant today. It is the same discipline as the knowledge state of an earlier chapter: a claim about what someone knows, scoped to a concept, with evidence behind it, and never a trait of the person. The mirror error also needs avoiding. There are genuine person-by-treatment interactions in the literature, involving for instance memory strategies, so the chapter does not claim that people never differ. Its principle is stronger than denial: the particular person-by-treatment relation has to be supported by evidence about that relation, and cannot be inferred from a global label.

A requirement is not a preference

Accessibility deserves its own boundary, because it is the place where treating a constraint as a score does the most harm. Suppose a system has evidence that a dense visual representation improves task speed in general, and Priya’s reader has asked for output that is compatible with text-only and keyboard use because of how she works. That is not a preference to be weighed against a speed advantage. It is a requirement.

preference           can influence a choice
accessibility need   constrains the choice

A preference can be traded against other considerations, with the trade made visible. An accessibility requirement is not traded away on the grounds that another form scored slightly better. It is honoured unconditionally, it is never inferred from behaviour or “matched”, and it sits alongside the preservation obligations of Part I as something that shapes what is acceptable before any optimising begins.

Preference is real without becoming authority

It remains true that “I prefer diagrams” is information, and it can reasonably change what the system does. What it must not do is harden into something else.

preference  →  evidence for adaptation
preference  →  not a permanent classification

“I prefer diagrams” may affect how a view is presented. It should not become “diagrams are how this person learns”, and it must certainly not become “never show this person prose”. A preference is stated in a context, for a kind of material, at a time, and it can change. An honest system keeps it as exactly that, a scoped piece of evidence that influences a choice and can be set aside when the material or the task calls for something else, with the reason visible. What one would expect from a fair comparison of the rival explanations is not a triumph for types at all: preference stays stable while performance varies with the task. People like diagrams throughout, prose wins on qualification-heavy material, tables win on exact comparison. That is no contradiction. It is two different objects, finally separated.

What replaces the type

If the type is the wrong unit, something else has to organise adaptation. The model is not a person reduced to a label but a decision that draws on several distinct inputs:

PERSON  +  TASK  +  PRIOR KNOWLEDGE  +  CURRENT CONDITION
+  ACCESS NEEDS  +  DECLARED POLICY
              ↓
     a representation decision

The task is what she is trying to do. Prior knowledge is what she knows about this thing, scoped and graded. Current condition is her situation now, declared or observable. Access needs are requirements. Declared policy is what she has told the system. Each has its own authority and its own way of changing, and none of them is a trait of the person. This is where several earlier chapters meet.

It also changes what the persistent state looks like. The system’s memory of Priya should be a collection of separately scoped claims, each with its origin:

declared    prefers a concise first view
declared    needs keyboard-accessible interaction
observed    often expands mathematical detail
knowledge   distributed consensus: strong
policy      interrupt only for active deployments

and not a profile:

PRIYA
  visual learner
  expert
  detail-oriented
  cautious

The second form is seductive because it is short and seems to explain a lot. It is the kind of simplification this part has to resist. Each word in it hides a different kind of claim, from a different kind of evidence, at a different level of confidence, and none of them is scoped. Collapsed into a label, they can no longer be questioned one at a time.

Where this leaves us

A person is not best treated as a stable representational type. Preference is real and is honoured, strategy and affordance explain most apparent fit without any type, expertise genuinely changes what helps but is scoped to a domain and changes over time, and accessibility needs are requirements that constrain the choice and are never traded against a score. Adaptation is assembled from separate sources, each with its own origin, scope and authority, and the system’s memory of a person should look like a set of those scoped claims, not a profile.

That leaves an obvious practical problem. If a person cannot safely be reduced to a type, then adaptation has to be built from many small pieces of evidence, accumulated over time: a correction here, a repeated choice there, a form that keeps working on one kind of task. Each piece is partial, each can go stale, and some will contradict each other. How should those pieces be remembered, revised, forgotten and reconciled?

References

• Pashler, H., McDaniel, M., Rohrer, D. & Bjork, R. (2008). Learning styles: Concepts and evidence. Psychological Science in the Public Interest, 9(3), 105–119.
• Clinton-Lisell, V. & Litzinger, C. (2024). Is it really a neuromyth? A meta-analysis of the learning styles matching hypothesis. Frontiers in Psychology, 15, 1428732. DOI 10.3389/fpsyg.2024.1428732.
• Hattie, J. & O’Leary, T. (2025). Learning styles, preferences, or strategies? An explanation for the resurgence of styles across many meta-analyses. Educational Psychology Review, 37, article 31. DOI 10.1007/s10648-025-10002-w. (Cited only for the point that matching research blurs styles, preferences and strategies.)
• Tetzlaff, L., Simonsmeier, B., Peters, T. & Brod, G. (2025). A cornerstone of adaptivity: A meta-analysis of the expertise reversal effect. Learning and Instruction, 98, 102142. DOI 10.1016/j.learninstruc.2025.102142. (Cited only for the direction of the effect.)
• Seufert, T., Schütze, M. & Brünken, R. (2009). Memory characteristics and modality in multimedia learning: An aptitude-treatment-interaction study. Learning and Instruction, 19(1), 28–42. DOI 10.1016/j.learninstruc.2008.01.002. (Cited only as an example that person-by-treatment interactions exist.)