How do different types of mental effort differ between scaffolded AI tutors, unrestricted chatbots, and ordinary study resources?

Why an open chatbot feels easier than a textbook

When students work physics problems with an unrestricted AI chatbot, they report lower mental effort than when they work from traditional textbook-style explanations [1]. The likely reason is that the chatbot can hand over the reasoning: it produces the next step, the formula, or the whole solution on request, so the student does not have to hold the problem in mind and search for a path forward. That is what makes it feel smooth.

The catch is that the smoothness is not free. The same comparison found that the reduced effort came with compromised depth of engagement [1]. In plain terms, students may finish the problem without having done the cognitive work that makes the method stick. This is an observation about how students engaged during the task, not proof that they learned less over time.

What a scaffolded tutor is supposed to do differently

A scaffolded tutor is built to give help in steps and then take it away. One design, CoMeT, escalates and fades its support based on what the learner actually does, rather than answering whatever is asked [2]. The idea is to keep the learner doing the part of the work they can manage while supplying just enough structure for the part they cannot.

That design was compared against an unrestricted assistant and against a tutor that could only ask questions, with cognitive load and frustration as the outcomes of interest [2]. The comparison is aimed at whether contingent help keeps effort at a productive level instead of either flooding students or leaving them stuck. Because this evidence comes from an abstract-level summary, the specific size and direction of the cognitive-load differences are not established here; what is clear is that the design question is about matching help to learner behavior.

Is lower effort always worse?

Not necessarily, and this is where the three tools separate. The tiered support system in the physics comparisons is a middle option: it offers structured help in layers rather than an open-ended chatbot or a static text [1][3]. Across these comparisons, the outcomes measured were cognitive load and affective-motivational factors, meaning how hard the task felt and how engaged or motivated students were, rather than achievement alone [1][3].

So the practical question is not simply which tool lowers effort most. A tool that lowers effort by doing the thinking for the student looks different from a tool that lowers effort by removing confusion while the student still does the reasoning. The chatbot evidence points to the first pattern; the scaffolded-tutor design is an attempt at the second [1][2].

What the comparisons do not yet settle

The evidence base here is thin in a specific way: the comparisons are described at the level of abstracts, and the outcomes center on cognitive load and motivation during problem solving [1][2][3]. That means we can say the tools differ in how much mental effort students expend and how engaged they feel, but we cannot yet say from this evidence which tool produces better long-term learning.

There is also a gap in coverage. The physics comparisons include an AI chatbot, tiered support, and textbook-style explanations, but they do not include a scaffolded AI tutor [1][3]. The adaptive-tutor study compares a scaffolded tutor with an unrestricted assistant and a question-only tutor, but it is a separate design [2]. Putting the two lines together is reasonable, but it is an inference rather than a head-to-head test of all three tools in one study.

Sources used in this answer

Related research questions