ChatGPT’s feedback credibility and students’ perceived learning: the role of agent-related social perceptions
Overview3:39CCAI
paperi.ai
0:00 / 0:00
Raúl Pérez‐López, Rubén González-Vallejo, Laura Lucia‐Palacios, Iguácel Melero‐Polo, Andreea Trifu
When ChatGPT gives writing advice, students may trust it not only because of what it says, but because the system seems capable and understanding. That trust is linked to how much learning they feel has happened.
This paper examines whether students’ agent-related social perceptions of ChatGPT are associated with the perceived credibility of its feedback in academic tasks and whether such credibility, in turn, is related to students’ perceived learning. It also investigates whether the relationship between feedback credibility and perceived learning varies according to students’ sensitivity to the quality of their academic work and their sensitivity to grade-based rewards. To address this, we develop a theoretical model grounded in social perception theory and prior research on students’ experiences with AI-based tools. The model is tested using Structural Equation Modelling (SEM) on data collected from 230 students who completed an academic writing task. The findings reveal that students’ agent-related social perceptions—particularly perceived competence and human-like empathy— are positively connected to the credibility students attribute to its feedback. In turn, perceived feedback credibility is interestingly related to students’ perceived learning in the context of academic writing. Furthermore, the positive link between credibility and perceived learning is weaker among students with higher gradebased reward sensitivity. This research contributes to the educational technology literature by advancing understanding of how students evaluate the credibility of ChatGPT-generated feedback and how such evaluations are associated with learning perceptions. It also shows that agent-related social perceptions play a significant role in students’ credibility judgements. Finally, the findings offer practical implications for educators and students seeking to integrate ChatGPT effectively into academic activities.
Transcript
When ChatGPT gives writing advice, students may trust it not only because of what it says, but because the system seems capable and understanding. That trust is linked to how much learning they feel has happened. Generative Artificial Intelligence technologies have been integrated into education and are described as important for enhancing academic skills and abilities, including language, critical thinking, analysis, and mathematical problem-solving.
Over recent years, generative AI has become integrated into education and important for developing academic skills, including language proficiency, critical thinking, analytical reasoning, and mathematical problem solving. ChatGPT is one of the most prominent generative artificial intelligence tools used by students in Western societies.
But its use in education remains under debate, because students increasingly rely on its feedback, often without checking the information sources and the reliability of the data behind it. The study asks whether students’ judgments about ChatGPT’s human-like characteristics and social abilities are connected to how credible they find its feedback.
It also asks whether credible feedback is related to students’ perceived learning, and whether that relationship changes with sensitivity to grades and academic quality. Students answered a survey about ChatGPT’s feedback credibility, their impressions of ChatGPT, and how much they felt they had learned.
The survey included judgments about warmth, competence, human-like empathy, credibility, perceived learning, sensitivity to grade-based rewards, and sensitivity to quality. The findings indicate that perceived competence and human-like empathy are positively related to students’ perceptions of feedback credibility.
In everyday terms, ChatGPT’s feedback worked more like advice from a trusted tutor: when students saw it as competent and able to provide correct answers, they were more likely to judge the feedback as credible. Empathetic responses were also associated with credibility, because students felt that ChatGPT understood their questions and could adopt their perspective.
ChatGPT’s feedback credibility was positively associated with perceived learning. The link was weaker among students who were more sensitive to grade-based rewards, while sensitivity to the quality of academic work did not show the same significant connection.
The three straight lines meet near the middle, then spread apart toward both ends, making the weakening relationship with grade-focused students easy to see. The research shows that students’ confidence in ChatGPT’s output for written academic activities varies according to their sensitivity to grade-based rewards.
It also examines how the credibility of ChatGPT’s feedback relates to students’ learning perceptions. Educators and students should critically evaluate the credibility of ChatGPT’s feedback, because generative artificial intelligence may provide incorrect, inaccurate, and sketchy recommendations.
Students should understand these limitations and use ChatGPT’s feedback to augment their own capabilities, while educators should look for detail and specific examples rather than relying on generic answers. Credible feedback is tied to stronger feelings of learning, but that connection is weaker for students driven more by grades.
ChatGPT can support learning, but its confident, human-like manner should never replace careful judgment.
A derivative work by Paperi · AI-generated script, voice and captions
· pages and figures unaltered
Made with Paperi.
Drop in a research PDF — get a narrated video walkthrough like this one,
with highlights that follow the narration. Free to start.
Medical records can help prevent disease and improve treatment, even when parts of a patient’s story are missing. But the way those blank spaces are handled can quietly change the conclusion doctors and researchers reach.Electronic health records can improve prevention, treatment, research, and even pandemic response, but the review finds that missing data are often handled inconsistently, leaving important uncertainty behind.
Yuanzhen Wang, Hongxin Zhang, Xing Fang, Xiaotong Cui, Wenxu Ning, Danzhi Wang, Fan Fan, Lei Shu
A device can reveal its encryption key without saying a word. Tiny electrical traces carry clues about the secret, but noise can bury them before anyone can read them.What if the best way to improve an AES side-channel attack is not a new attack at all, but a smarter way to clean the measurements first? This paper uses singular value decomposition to keep useful leakage while suppressing noise.
Maurizio Giordano, Lucia Maddalena, Mario Manzo, Mario Rosario Guarracino
An artificial-intelligence system can make sense of a complicated network, yet an attacker may not need to touch the cases it must judge. Altering its training examples can be enough to make the system fail.What happens when an attacker does not need to touch a model directly, but only removes selected edges from the graphs used to train it? This study shows how graph embeddings can be tested under exactly that kind of poisoning attack.
3:12
