The promises of artificial intelligence and efficiency threaten the core of learning

Digitalisation has been introduced into higher education by promising efficiency and flexibility. Technology has been expected to make teachers’ workloads lighter (Allen & Seaman 2017; Du Plooy et al. 2024), personalise learning experience (Hooshyar et al. 2024), improve accessibility (Briggs et al. 2024), and even boost learning motivation through gamification (Li et al. 2024).

Yet realising these promises has proved more complex than technological optimism has suggested. At the same time, these narratives have obscured a more fundamental question: what, ultimately, is learning in higher education, and what should it enable?

Technology has been regarded as a central driver of higher education reform (e.g., De Nito et al. 2023). Its promises are built on a discourse of efficiency. The logic of this discourse relies on a cause–and–effect thinking, in which technology sets the terms of action. At its foundation lies data-drivenness which transforms human interaction and learning into measurable raw material.

This is followed by administrative optimisation, which harnesses accumulated data to streamline production processes and eliminate waste. The result is a technological imperative in which measurable efficiency displaces qualitative and pedagogical values as the primary steering mechanism of the university (Komljenovic et al. 2025; Williamson 2017).

The question that follows: What is learning in higher education, and what should it make possible?

Critical scholarship has argued that the digitalisation of higher education is neither neutral nor inevitable but political. Technologies are socially constructed and negotiated rather than carrying predetermined characteristics, and their use in universities is shaped by social conflicts and competing interests. These negotiations are tied to societal power relations and economic conditions, and therefore the technologisation of higher education serves some interests more than others and shapes social reality unevenly. (Selwyn 2010.)

This problem is especially salient in the case of generative AI based on large language models. LLMs are trained on vast corpora, and they can produce human-like text, answer questions, and perform a range of language tasks (e.g., Kasneci et al. 2023). Hereafter, we use the term AI to refer specifically to generative AI applications built on large language models.

AI-driven change may temporarily improve academic performance, yet it simultaneously risks narrowing the development of competencies required in new work environments (Panadero & Broadbent 2025). When students use AI to complete their assignments, the technology can extend learning, but just as easily, it can replace it altogether. The pen extended the possibilities of writing, and the calculator sped up computation, but both still required thinking. AI offers a tempting shortcut that can bypass the very process of thinking (Purser 2025).

In this article, we examine how technological disruption and AI challenge critical thinking and traditional study practices. We argue that, in an era that idealises efficiency, higher education institutions must ensure space and porosity for students’ identity work. Only then can we ensure that critical and responsible learnerhood, which is grounded in trust between educators and students, endures in the age of AI.

AI’s epistemic limitations and demands

Despite advances in AI, learning still requires effort, persistence, and critical thinking. By critical thinking we mean the capacity and willingness to analyse and evaluate situations, which involves cognitive effort, reflection, and questioning prior knowledge and assumptions (Larson et al. 2024).

Critical thinking presupposes an understanding of how knowledge is constructed and assessed in different disciplines, which is emphasised when AI becomes part of the learning process. AI models generate text based on statistical probabilities; they do not verify their accuracy against trustworthy sources (Zhang et al. 2025). Validating AI-generated content, therefore, demands a deep understanding of how knowledge is produced and warranted within a given field (Potkalitsky 2025).

Leveraging AI requires users to engage in critical information evaluation, which creates a pedagogical tension in higher education. Students are expected to take responsibility for texts whose expert evaluation may still exceed their competence. At the same time, working with AI can weaken the human relationships in which critical thinking typically develops (Flenady & Sparrow 2025). Critical thinking emerges gradually, through sustained engagement with disciplinary practices of knowledge formation.

Learning is a multifaceted process that requires cognitive, emotional, motivational, and behavioural work. AI is increasingly implicated in this process (see Panadero & Broadbent 2025). It can render students passive by providing ready-made answers too readily. If students mistakenly regard AI-generated responses as objective and coherent (Larson et al. 2024), their development of expertise is jeopardised.

Conversely, using AI for routine and analytical tasks can support learning, as long as technology is enlisted as a scaffold for cognition rather than a substitute for it (Panadero & Broadbent 2025).

Porous time as a condition for learning and expertise

To avoid becoming trapped by AI-enabled efficiency in higher education, we need porous time. Porous time foregrounds the permeability and unpredictability of time, emphasising that time cannot be entirely governed through rational planning or segmented into neatly bounded units. The notion underscores how time flows flexibly through social structures, even when those structures are not designed to recognise such complexity (Mazmanian et al. 2015). For learning, this means that learners can pause to reflect, connect new knowledge to their experiences, and integrate it into their developing sense of self.

Such learning spaces may foster more profound understanding, stronger memory traces, and the emergence of creative insight. The role of the university educator is not merely to transmit knowledge but to guide students in developing metacognitive skills, in other words, how to learn (Schraw et al. 2006). Above all, this means that students retain autonomy regardless of the conveniences offered by AI. The key task of an educator in higher education is to enable learning process and to support students in cultivating their metacognitive capacities.

Learning no longer entails only mastering content; it also involves comparing one’s own thinking with AI-generated output. The most skillful students use AI to interrogate their own knowledge and the processes by which knowledge is produced. Here, evaluative judgement is continually developing, contextual, reflective, and grounded in ethical commitment (Bearman et al. 2025).

Higher education is, therefore, at a turning point as AI generates answers that appear authoritative. Counterintuitively, this calls into question existing efficiency ideals and elevates the importance of critical thinking, substantive disciplinary knowledge, and the development of expertise. This, in turn, requires time, deliberation, and above all, the courage to pause and reflect on why one is learning and why it matters.

Trust in teacher–student relationships

In the most critical commentaries, AI is feared to destroy universities and deep learning. When both teachers and students lean heavily on AI, a singular pursuit of efficiency can erode learning. Degrees may then cease to indicate genuine competence (Purser 2025).

This risk must be taken seriously. To avoid it, we need to understand how AI affects trust, both in interpersonal relationships and in relationships with technology.

In this text, we rely on the definition of interpersonal trust as a student’s willingness to be vulnerable to a university teacher’s actions. This willingness is based on the student’s assessment of the teacher’s trustworthiness, which is typically a function of ability, benevolence, and integrity (Mayer et al. 1995). In educational settings, trust crystallises as confidence that the teacher is committed to advancing the student’s interests and does so competently and with honesty and openness (Hoy & Tschannen-Moran 1999; Luo 2025).

The use of AI introduces a challenging asymmetry of transparency: students are expected to be open in ways that teachers may not be (Luo 2025). A lack of transparency on the part of teachers can be perceived as hypocritical, clashing with students’ expectations for human-centred teaching (Hill 2025). The danger is a loss of trust between teachers and students.

A weak trust climate may hinder students from using AI because they fear negative consequences in assessment (Luo 2025). Teachers should communicate their own AI practices and pedagogically justify any limitations on use. In this way, students learn to recognise situations in which the efficiency afforded by AI does not substitute for learning.

Trust in AI

According to Glikson and Woolley (2020), trust in AI is both emotional and cognitive. Cognitive trust concerns the willingness to rely on AI-generated information and rests on the technology’s intelligibility and the consistency of its outputs. Because AI’s answers vary by context and user, such trust is not stable. This volatility can dampen the inclination to rely on AI, underscoring the importance of critical thinking and learning when using technology.

Emotional trust in technology becomes salient, especially in tasks that demand interaction. A key factor is the humanisation of technology: the more human-like AI appears, the more it can elicit trust, while simultaneously raising expectations of its competence (Glikson & Woolley 2020). For example, an encouraging tone from an AI system can strengthen emotional trust, even when users remain sceptical about the accuracy of the content. Consequently, students and teachers may use AI on the basis of emotional trust, even when their cognitive trust in its correctness is relatively lower.

Whereas high trust in AI can dampen critical thinking and reduce cognitive effort higher self-confidence is associated with more critical thinking , boosting both. The task in higher education is to find an appropriate balance: low confidence in one’s abilities may lead to excessive reliance on AI, thereby weakening independent problem-solving. Teaching should thus boost students’ professional self-confidence and provide tools for assessing AI’s reliability (e.g., Lee et al. 2025).

The core mission of teaching and guidance has shifted in a consequential way: the work is no longer only to build students’ awareness of, and confidence in, what they know; it is also to strengthen their trust in their own thinking processes.

Higher education as a site of identity formation

Universities are not merely institutions for credentialing or knowledge acquisition; they are key transitional contexts in which students reconstruct their relationship to self and world. As MacFarlane (2018) suggests, entry into higher education resembles a rite of passage: students must refine their values and confront the fundamental question, ‘Who am I?’ (Baxter Magolda 2007). Such identity work is not a single epiphany, but an ongoing negotiation conducted through cognitive reflection and social interaction (Alvesson & Willmott 2002; Caza et al. 2018).

This is precisely where the relationship to AI becomes meaningful. AI can deliver immediate efficiency and swiftly formulated responses, but identity construction requires time, tolerance for incompleteness, and porous spaces. The ostensibly inefficient moments of campus life, such as student events, meetings in the coffee queue, hallway conversations, and the public trying-out of half-formed ideas, are often the very situations in which meaningful identity work occurs. When these spaces are systematically filled with performance-driven efficiency, students’ opportunities to grow into independent agents may be narrowed.

The strengthening of student identity and self-confidence appears as mutually reinforcing developmental processes (Luyckx et al. 2013). Such self-confidence is needed for critical (Lee et al. 2025) and purposive use of AI. If students do not recognise their own strengths, the temptation to outsource thinking to AI may grow.

For this reason, the role of university teachers is not limited to ensuring mastery of content; they also act as enablers of identity work. By linking learning tasks to students’ lived worlds, making thinking processes visible, and nurturing community, teachers can protect the spaces of human growth that technological efficiency cannot replace.

Learning as a collective endeavour instead of the logic of efficiency

Reclaiming teaching and learning as relational practices entails questioning the logic of efficiency and embracing both the stickiness of learning (see Alanko-Turunen & Kaukinen 2025) and the pedagogy of discomfort. This applies equally to the use of AI.

The problem is not AI per se, but the extent to which students use it without pedagogical framing or reflection (Panadero & Broadbent 2025). Rather than a simple allowed/forbidden dichotomy, what is needed is an ongoing dialogue about what is meaningful for students’ futures (Bearman et al. 2025), precisely the relational, porous negotiation that sticky learning demands.

A central task for university teachers is to support students’ developing professional identities so they can participate ethically and sustainably in an AI-shaped world of work. This requires designing learning experiences as a trust-grounded collective endeavour, not merely as the completion of tasks, and reconsidering how teaching and learning time are structured so that students become actively invested in their own and each other’s learning.

Relational responsibility is shared. For students, this calls for more than merely meeting contemporary AI standards: it requires the capacity to assume new responsibilities and to act ethically in complex situations (see Hamilton et al. 2025). Such autonomous agency in learning does not emerge from individual obligation alone, but rather in porous spaces without constant technological efficiency requirements.

Picture: Haaga-Helia