Summary: Researchers observed 90 adult learners attempting to interpret object-placement instructions given in Swahili by a humanoid robot. The study revealed a clear temporal trade-off: highly detailed, personalized feedback increased cognitive load and reduced the effectiveness of immediate, task-focused hints on the very next attempt, yet produced stronger overall task performance across the session.
These results suggest robotic tutors must balance personalization with moment-to-moment assessment of a learner’s cognitive and emotional state, knowing when brief guidance is more effective than detailed instruction.
Key Facts
- Swahili spatial puzzle task: Ninety adult participants placed everyday objects (bottle, cup, book) in correct room locations based on spoken instructions from a humanoid robot in Swahili, a language unfamiliar to them, so learners had to progressively decipher the meanings to solve the task.
- Three feedback approaches compared: The study evaluated (1) fixed, non-adaptive feedback delivered after errors; (2) performance-and-enjoyment-adaptive feedback that adjusted frequency to recent performance and self-reported enjoyment; and (3) personalized-adaptive feedback that also tailored content to a learner’s past errors and prior steps.
- Immediate versus overall performance trade-off: Personalized content improved overall task mastery across the experiment but temporarily diminished the immediate usefulness of task-focused hints for the learner’s very next response, likely because the added specificity increased processing demands.
- Moderation by cognitive ability: Learners with higher cognitive ability benefited less from immediate task-focused hints, appearing to generate their own corrections; extra guidance sometimes distracted rather than helped them.
- Affect and attention: Participants reporting higher situational boredom gained more from concise, task-focused hints, which served to re-engage their attention and improve subsequent performance.
Source: TUB
Scenario: A learner stands in a room with a bottle, a cup, and a book. The goal is to determine step by step where each object belongs—under the table, on the chair, or elsewhere—based on a humanoid robot’s spoken instructions in Swahili. Because the learners did not know the language, they needed to infer meaning across multiple trials.
After every placement the robot provided feedback. Sometimes it simply indicated the placement was incorrect; at other times it offered a hint to guide the next move or prompted the learner to reflect on their approach. The study, titled “Real-time cognitive-affective dynamics of failure feedback in a technology-based learning task,” came from the Cluster of Excellence Science of Intelligence (SCIoI) in Berlin and appeared in Communications Psychology. Authors Helene Ackermann, Anna L. Lange, Hanna Dumont, Verena V. Hafner, and Rebecca Lazarides analyzed how automated feedback delivered by a humanoid robot influenced learners’ momentary processing and overall task success.
As humanoid robots and AI tutors are considered for classrooms and workplaces, the study addresses a practical question: when and how does robotic feedback actually support learning? The findings are nuanced: automated hints help learners recover from mistakes, but more personalized feedback is not always better in the immediate aftermath of an error.
“Automated feedback can support learning after mistakes,” says Helene Ackermann. “But help must match the moment: right after an error, more information is not always better.”
The timing of help
The research compared three conditions. In the fixed guidance condition, the robot delivered additional feedback after every mistake without adapting to the learner. In the basic-adaptive condition, feedback frequency adjusted to recent performance and enjoyment but remained content-generic. In the personalized-adaptive condition, feedback frequency adapted and the content was tailored to the learner’s specific errors and prior attempts.
Personalized feedback seems intuitively best because it references previous mistakes and steps—for example, reminding a learner they had already tried a particular placement. Yet the results showed complexity: personalized messages tended to be longer and richer in detail, which appeared to increase cognitive load immediately after an error. That extra processing demand reduced the effectiveness of concise task-focused hints on the learner’s very next attempt.
Despite this short-term slowdown, personalized feedback was associated with improved performance over the entire task, indicating a trade-off between immediate responsiveness and long-term learning gains.
“Personalized feedback wasn’t simply good or bad,” says Anna Lange. “Its effect depended on timing: it could hinder the next move right after a mistake but still support better performance across the task.”
Robots need to read the situation, not just the error
The study also showed that the same feedback can affect learners differently. Higher cognitive-ability participants benefited less from immediate task-focused hints, likely because they could self-correct without additional prompts; extra information sometimes became distracting. Conversely, learners who reported more situational boredom benefited notably from concise, task-oriented feedback that helped re-engage attention.
Altogether, the findings emphasize that effective robotic support depends on both the content and timing of information, and on the learner’s momentary cognitive and emotional state. Systems that combine personalization with sensitive, real-time sensing of those states can optimize when to offer short hints and when to provide richer guidance.
“Personalization is not the end goal,” says Rebecca Lazarides. “The challenge is dynamic: intelligent systems must blend tailored content with situational awareness and timing.”
What robot tutors still need to learn
As robots become part of public conversations about education and assistance, this study is a reminder that more support is not always better. The findings do not argue against personalized feedback—indeed, it produced the best overall outcomes—but they caution that detailed support can raise cognitive load in the immediate aftermath of errors. Designers of human-robot interactions should therefore balance personalization with brevity and build systems that adapt to learners’ situational cognitive and affective states.
For SCIoI, the study advances understanding of intelligent interaction: an effective robot partner must not only know what to say but also when to say less.
Key Questions Answered:
A: Personalized feedback included specific details about a learner’s prior attempts and error history. Immediately after a mistake, processing this additional detail increased cognitive load, making it harder to apply the hint on the very next attempt.
A: It helped overall. Although it temporarily reduced the effectiveness of immediate task-focused hints, personalized feedback led to the highest cumulative task success across the experiment by supporting deeper understanding over time.
A: Learners who reported higher situational boredom benefited more from concise, task-focused hints. Those hints acted as an attentional anchor and helped re-engage participants in the task.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- Journal paper reviewed in full.
- Additional context provided by staff editors.
About this robotics and neurotech research news
Author: Maria Ott
Source: TUB
Contact: Maria Ott – TUB
Image credit: Neuroscience News
Original research: Open access. “Real-time cognitive-affective dynamics of failure feedback in a technology-based learning task” by Helene Ackermann, Anna L. Lange, Hanna Dumont, Verena V. Hafner & Rebecca Lazarides. DOI: 10.1038/s44271-026-00487-8
Abstract
Real-time cognitive-affective dynamics of failure feedback in a technology-based learning task
As technology-based learning environments increasingly use automated feedback, it is vital to understand how learners process that feedback in real time. This study examined how cognitive and metacognitive failure feedback delivered by a humanoid robot affected performance and how those effects were moderated by feedback design and individual learner characteristics.
Ninety adults (ages 18–59, mean age 29.53; 61 female, 27 male, 2 diverse) completed a spatial learning task under three conditions: (1) fixed guidance with fixed-frequency, content-generic feedback; (2) basic-adaptive with frequency-adaptive but content-generic feedback; and (3) personalized-adaptive with frequency-adaptive, content-personalized feedback tailored to specific errors and prior steps.
Using a three-level generalized path model (trials nested within time blocks within learners), the analysis found that cognitive and metacognitive failure feedback generally increased the likelihood of a correct subsequent response. Relative to fixed guidance, frequency-adaptive feedback showed no significant moderation. Content-personalized feedback reduced the immediate effectiveness of cognitive failure feedback but improved overall performance compared to content-generic feedback. Across conditions, higher cognitive ability was associated with smaller benefits from feedback, while higher momentary on-task boredom was associated with larger benefits.
These findings highlight that the effectiveness of automated failure feedback depends on both its design and learners’ situational cognitive and emotional states, demonstrating the value of temporally sensitive, context-aware feedback strategies.