Understanding the place and value of GenAI feedback: a recognition-based framework

Feedback as information: “feedback is information that includes all or several components: students’ current state, information about where they are, where they are headed and how to get there, and can be presented by different agents (i.e., peer, teacher, self, task itself, computer). This information is expected to have a stronger effect on performance and learning if it encourages students to engage in active processing.” (Lipnevich and Panadero Citation2021, p25)

Feedback as process: Effective feedback is … “a process in which learners make sense of information about their performance and use it to enhance the quality of their work or learning strategies.” (Henderson, Ryan, and Phillips Citation2019, p1402)

Feedback messages: form and function

Feedback information has long been acknowledged as taking different forms, and originating from different sources, despite the teacher or educator archetypically considered the arbiter of judgement, and therefore the generator of information, and the student seen as a willing recipient of information. Hattie and Timperley (Citation2007) review highlighted feedback as a powerful influence on learning and achievement. Their model of feedback – where information can focus on the task, process, self-regulation, or self – also characterised the content of effective feedback messages around three prompting questions: “Where am I going? How am I going? Where to next?” (p88). Beyond the content of the message itself, in their integrative review of feedback models and typologies, Panadero and Lipnevich (Citation2022) identified four additional dimensions of feedback information: how feedback is implemented, student characteristics, the context, and agents involved. These dimensions are important to consider in light of the possibilities for GenAI feedback.

The intended functions of feedback encompass contextual and implementation dimensions, as well as influence the content of the message itself. While binary distinctions such as formal/informal or formative/summative might be helpful for educators to imagine and plan for feedback, they are generally unhelpful in understanding what unfolds in practice, since the importance and impact of feedback messages greatly depends on a range of contextual factors and the perceptions and situation of the individuals for whom the message is intended. That said, informal feedback might be considered as more likely to occur during classroom interactions (Sambell Citation2011) while formal feedback is more systematically integrated into curricula and typically associated with assessment tasks (Värlander Citation2008). Formative feedback is anything that might support further growth and development, while summative feedback tends to be associated with judgement activities which ‘counts’ – towards marks, grades, or progression.

The form or medium of the feedback message can then be considered, also in relation to the feedback agent. While written text, audio, video, and haptic (involving tactile sensations, such as vibrations or pressure) feedback are commonly considered with respect to human feedback agents (who could be educators, peers, selves, or knowledgeable others beyond educational systems), there is also the possibility that these might be generated through computer systems – some of which might be commonly expected within particular professional or disciplinary environments (e.g. automated computer feedback when debugging software is run; embodied feedback as part of clinical skills teaching in the health professions).

It is not controversial that generative AI can generate feedback messages for various functions and in various forms. Students have already identified that GenAI can provide feedback information, with documented examples of students providing a system like ChatGPT a copy of an assessment task and asking for feedback prior to them submitting it formally (Johnston et al. Citation2024). GenAI is also increasingly being integrated within formal education to do this (Kiyasseh et al. Citation2023; Xu et al. Citation2023; Taylor and Marino Citation2024). Studies also exist showing that AI is capable of generating formative or constructive feedback of various kinds (Banihashem et al. Citation2024; Guo et al. Citation2024; Morris et al. Citation2024; Pang, Kootsookos, and Cheng Citation2024, Jürgensmeier and Skiera Citation2024; Yin et al. Citation2024).

Feedback interlocutors: vulnerability and trust

While GenAI is able to generate feedback messages, its role as an interlocutor in feedback processes requires further consideration. Contemporary conceptualisations of feedback acknowledge a broad and nuanced landscape of feedback processes which extend beyond individual interactions between a student and educator. This process perspective of feedback considers feedback to have occurred when the learner or student has done something with the information they received. Whilst commonly thought of as that which happens between a learner and another person, processes of educational feedback can also happen between human and non-humans – for instance, where information might be generated in the course of working with a computer (Esterhazy Citation2018). Students may therefore assign a hierarchy of trust and value to feedback processes, on the basis of the interlocutor’s qualities, beyond the content of the message received (Tai et al. Citation2017; Zhou, Zheng, and Tai Citation2020). Importantly, discussions of effective feedback processes are often framed around the establishment of trusting and respectful relations between students and others, in connection to the affective and relational dimensions of feedback.

The affective dimensions of feedback are important to student experiences of feedback. Students may encounter discomfort in receiving critical feedback from teachers, especially when anxiety and challenges to self-esteem are often heightened (Shields Citation2015). Feedback processes often invoke strong emotions or threats to self-esteem, so handling them sensitively enhances the potential uptake of key messages (Carless and Winstone Citation2023). Given the prominence of negative affect, feedback processes are likely to be enhanced when relational support is offered through emotional sensitivity, empathy and trust (Steen-Utheim and Wittek Citation2017). Carless suggests that “trusting virtues such as empathy, tact and a genuine willingness to listen are ways in which positive feedback messages can flourish and more critical ones be softened.” (Carless Citation2012, p90). Carless and Boud (Citation2018) also highlights the capacity to manage affect as part of feedback literacy.

Sharing feedback experiences and foregrounding vulnerability – a kind of “ceding of power” in the classroom – has been highlighted as a helpful quality in establishing effective feedback relationships. Teachers’ offering of their own experience of feedback from peer review and from student course evaluations have been suggested as ways to manage the affective component of feedback practices (Carless and Boud Citation2018, Gravett et al. Citation2020, Molloy and Bearman Citation2019). This helps to establish what Carless and Winstone term a “partnership in feedback,” in which students feel a reciprocal responsibility to the teacher, taking on “increased responsibilities” and opening up dialogic interaction with teachers (Carless and Winstone Citation2023). These strategies for managing affect acknowledge the deeply relational nature of effective feedback process. In feedback, both parties – both the teacher(s) and the student(s) – are vulnerable to the judgment of the other, the student explicitly, insofar as it is their work being evaluated; the teacher implicitly, insofar as their capacity to exercise good judgment is under scrutiny. This vulnerability, as Carless (Citation2012) following Tschannen-Moran (Citation2004) has argued, is not a deficiency to be overcome, but rather the condition of genuinely trusting relationships between teacher and student: to trust is to be willing to be vulnerable, an “investment of faith” in the other (Carless Citation2012, p. 91). And, as Carless demonstrates, trusting relations between teachers and students is essential to the construction of genuinely productive, dialogic feedback processes (Carless Citation2012).

Recent literature on feedback has foregrounded human participants in processes, and has relatively less to say about how students might trust computers or inanimate objects as partners in feedback. There has been an unarticulated expectation that the interaction between learner and interlocutor (if there is one) takes place within a developing relationship. Accounts of peer feedback suggest that the qualities of the interlocutor can vastly alter the student experience of feedback (Panadero et al. Citation2023). For example, recently, Zhou et al. (Citation2021) argued that effective feedback is dependent on respect, focussing on “evaluative” and “care respect” as two important dimensions, not just for softening the emotional impact of feedback, but to ensure that feedback messages are interpreted and used productively. Moreover, in line with the discussion of mutual vulnerability above, respect is engendered through acknowledging and identifying teachers’ imperfection and incompleteness, perhaps by foregrounding their own experiences of peer and student feedback.

Arguments which proffer GenAI as a replacement for human feedback interlocutors, with the only difference being that AI systems can be scaled up to reach more students, ignore this relational aspect of feedback. This elision of the qualitative difference between human and AI feedback, we suggest, is deeply problematic and serves to obscure both the ethical and instrumental limitations of AI feedback systems. In the following section, we therefore turn to the philosophical literature on mutual recognition to extend our understanding and clarify the connections between affect, vulnerability, relationality, and respect, to identify what GenAI can and cannot do in feedback.

Recognition and feedback

Our introduction of Brandom’s and Honneth’s respective post-Hegelian social philosophies clarifies the dependency on effective feedback on appropriately mutually recognitive relations between teacher and student, in two interrelated ways:

Firstly, while the higher education literature on feedback does note the vulnerability of students in the feedback process (see Section 1.2 above), the above philosophical discussion shows that this vulnerability in question is a fundamental one, insofar as one’s social role and its corollary entitlements and commitments is ultimately dependent on recognition by others. One is not, in other words, who one is and then recognised as such; one is only who one is through recognition (Deranty and DePaul University 2011). Student vulnerability in feedback thus should not be an afterthought calling for a “soft touch,” rather, respecting that vulnerability requires an awareness of how, in a sense, the student themselves is at stake in feedback. They are, in one sense, not a student at all until they are recognised as such by another, nor can they understand themselves to be meeting (or failing to meet) disciplinary norms without the establishment of mutually recognitive relations.

Secondly, the fundamental nature of recognition helps to draw out more clearly the mutual vulnerability and thus mutual responsibility of teacher and student; both are dependent on the other for recognition of their social role and its corollary entitlements and obligations. This helps to sharpen the claims treated above, namely, that a genuinely dialogic feedback practice is dependent on each party recognising the other in their vulnerability, insofar as such mutual recognition gives rise to a responsibility to treat the other party with respect. To put it directly: in a respectful and trusting relationship, one recognises the other as fundamentally like oneself, that is, a being whose self-understanding and self-esteem is dependent on and vulnerable to the judgments of others. In the pedagogical context, one ought to recognise the other as, like oneself, desirous to be recognised as possessing certain capacities but doing so as “ordinary, imperfect, limited, incomplete, and always under construction” (Dillon Citation1992: 121).

Recognitive feedback

As we conceive it, recognition-capable feedback, or “recognitive feedback,” is characterized by potentially mutually recognitive relations between interlocuters, namely, between the feedback provider and the recipient. This form of feedback is not reducible to mere information transfer or performance evaluation, nor is it dependent to the quality or what we might term the authority of the information in isolation; it is a process through which both student and educator identities are affirmed and developed. Central to recognitive feedback is the mutual vulnerability of both parties to the judgment of the other. This vulnerability is not a weakness to be overcome but rather a necessary condition for the establishment of trusting and respectful relationships and, consequently, for effective feedback. The educator, in providing feedback, opens themselves to the student’s judgment of their expertise and pedagogical skills. Simultaneously, the student, in receiving and engaging with feedback, exposes their developing academic identity to the educator’s assessment. This means that students develop their academic identities through being recognized as capable, developing scholars by their educators. Conversely, educators reaffirm and develop their professional identities through student engagement with and recognition of their feedback. Consequently, when their relationship with students is challenged, for example as a result of emerging GenAI technology (Corbin et al., Citation2025), it can result in challenges to teachers very identity as teachers.

Extra-recognitive feedback

In contrast, recognition-incapable or “extra-recognitive” feedback lacks this essential element of mutual recognition. While it may provide accurate information or content “personalized” to the student’s previous learning and assessment, it fails to engage in the deeper process of recognizing the learner as a developing scholar. Extra-recognitive feedback is primarily a unidirectional transmission of information, lacking the reciprocal nature that characterizes recognition-capable feedback. In other words, extra-recognitive feedback may achieve a surface-level personalization by referencing specific details of a student’s work, but it is unlikely to comprise a deeper understanding of the student’s unique learning context and trajectory.

Diagnostically speaking, where extra-recognitive feedback tries to replace recognitive feedback, a student’s desire to have their effort and developing capacities recognised by another human being who shares in that desire is thwarted. A GenAI system might provide a semblance of achievement recognition and respect for the vulnerability of the student to judgment, by, say, congratulating a student on the work they have done in order to submit a given piece of assessment on time. But such a congratulation, however nicely worded, lacks the recognitive significance expressed by another human being who genuinely shares the experience of working hard, challenging oneself, expanding one’s knowledge, and so on. At the same time, however, we suggest that extra-recognitive feedback, if properly instituted, may be a pedagogically valuable supplement to recognitive feedback processes. In the following section, we set out the limitations and benefits of AI systems that this novel distinction brings into view.

The recognitive framework in practice

We suggest that the distinction between recognitive and extra-recognitive feedback offers significant analytical, diagnostic, and practical advantages for understanding both traditional and emerging GenAI feedback practices. By making explicit what was previously implicit in feedback processes, this framework allows us to better theorize existing feedback models while also appreciating more clearly both the limitations and unique affordances of GenAI feedback systems. Here, by way of unpacking our framework, we explore its application in several practical scenarios.