Co‑constructing Visions for the Future of Teaching

11. Hybrid human-AI collaboration in education as future scenario

Abstract

Inge Molenaar is professor of Education and AI at the at Radboud University and director of the Dutch National Education Lab AI (NOLAI). She envisions a future in which AI functionalities are designed and developed specifically from an educational perspective and explores how AI can be integrated into educational practices, based on different modes of teacher-AI collaboration.

An expert perspective by Inge Molenaar, professor of Education and AI at the at Radboud University and director of the Dutch National Education Lab AI (NOLAI)

In this short future vision essay, I will explore possible futures of education supported by AI. The role of AI that I envision is one in which the technology is specially designed for education. This “educational AI” is informed by our scientific knowledge about teaching and learning and designed and developed in close collaboration with teachers. This leads to AI applications in which the functionalities are designed and developed from an educational perspective. In this way, novel educational practices unfeasible without AI technologies are created that complement and augment human teachers (Molenaar, 2024[1]).

We speak of complementation when existing teaching practices are enhanced by using AI, which allows teachers to improve existing pedagogical instructional practices. In contrast, when AI augments teachers, this allows for novel pedagogical and instructional practices, which are only possible when teachers and AI collaborate. Both complementation and augmentation can be typified as human-centred processes and stand in contrast to teacher replacement, which is technology-centred.

There are at least two critical factors for the realisation of this future. First, the seamless connection between existing teachers' practices and the functionality of AI in teacher-AI collaboration. This is crucial for the successful deployment and sustained usage of AI applications in classrooms. Second, teacher complementation and augmentation scenarios should be informed by scientific knowledge on teaching and learning. In this way, teacher-AI collaboration will be embodied in evidence-informed innovations which enter the classroom to enhance future teaching and learning.

However, we are only partially on track for this future. Research on teacher-AI collaboration is still in its infancy (Cukurova, Kent and Luckin, 2019[2]; Vaccaro, Almaatouq and Malone, 2024[3]). Much research and policy attention is focused on the effectiveness of AI solutions for improving student learning. This evidences a profound underestimation of teachers' important role in deploying AI in education (Tammets and Ley, 2023[4]) (Ley et al., 2025[5]). The focus on the effectiveness of AI technologies stems from the assumption that the AI alone is responsible for effects on learning, implicitly assuming that teachers may be replaced by the AI (Molenaar, 2021[6]). However, the teacher and AI function as a combined social-technical system in most AI classroom applications (Cukurova, 2024[7]). Hence, our attention should be focused on the effectiveness of the combined teacher-AI system. Although there is evidence that teacher-AI combinations are more effective than AI alone (Aleven et al., 2023[8]), research on teacher-AI collaboration and how to design, develop and implement it is still scarce.

Similarly, using scientific insights on learning and teaching in AI design and development is still uncommon. Research shows that only 49% of learner-facing AI solutions include scientific knowledge about human learning (Topali et al., 2025[9]), and for teacher-facing AI solutions, only 33% are designed and developed based on scientific understanding of teaching (Costache, under review[10]). This indicates there is a considerable gap between our current scientific insights into how people learn and the functioning of AI in educational technologies.

I envision a future in which we advance in understanding teacher-AI collaboration, the alignment between scientific knowledge on learning and teaching, and the functionalities of AI in educational solutions (Molenaar, 2022[11]; Molenaar, 2024[1]). This essay will explore different types of teacher-AI collaboration, the challenges in designing such systems, and the steps required to ensure responsible AI integration in education. I will emphasize how education, research, industry, and policy can collaborate to make informed choices for the design, development, and deployment of human-AI collaboration. Processes of co-creation and co-implementation are outlined as elements towards a future with sustainable and responsible application of AI in education.

Perspectives on Human-AI collaboration in education

As the introduction indicates, the technology-centric approach to designing, developing, and deploying AI is still mostly dominant (Shneidermann, 2022[12]). This perspective envisions that AI will replace humans. An example is the self-driving car in which the AI takes over driving from humans. Yet, in education, only 6% of teaching tasks are expected to be replaced by AI, and 22% of teaching tasks are classified as bottleneck tasks, i.e., tasks that are highly unlikely to be performed by AI in the near future (Lassébie and Quintini, 2022[13]). Hence, there will be ample space for human-AI collaboration within the teaching profession, 72% of the tasks could entail a task division between teachers and AI.

Therefore, successful teacher-AI collaboration is a critical element in the future of AI in education. Hybrid Intelligence aims to research and develop intelligent systems that augment rather than replace human intelligence (Akata et al., 2020[14]). These combined systems leverage human strengths and compensate for human weaknesses. Humans and AI are regarded as equal team members solving tasks in cooperation (Siemon et al., 2017[15]). Hybrid intelligent systems elevate the complementary strengths of human intelligence and AI and enhance the combined system (Dellermann et al., 2019[16]). Such collaboration between human and artificial intelligence as a social, technological system is an essential element of the future of AI in education (Cukurova, 2024[7]; Molenaar, 2022[11]).

Yet, research on hybrid intelligence is still in an early stage, and there are numerous challenges in designing and developing such combined systems. A recent meta-review that investigated human-AI combinations compared to AI and human performance alone showed that human-AI combinations often performed worse than the best of humans or AI alone (Vaccaro et al. 2024). Especially with respect to decision-making, performance losses of combined systems were found, whereas they did better for content creation. Interestingly, when humans outperformed AI alone, combined systems did better, but when AI outperformed humans, combined systems did worse. Results could be partially explained by humans delegating the wrong tasks to AI. In light of this future vision, it is important to note that this meta-review did not include any studies in education, and the effects of combined systems in education have hardly been researched so far.

The finding that AI can augment humans and improve performance in systems where humans outperform AI suggests the potential for human-AI collaboration in complex social-technical ecosystems such as educational practice. These benefits are expected, especially in cases where humans and AI excel in different tasks. Teachers and AI often have distinct roles in classrooms and outperform each other on different subtasks. In the next section, I will outline examples of educational tasks with distinct tasks for teachers and AI, and illustrate different types of teacher-AI collaboration. After having explored the broader perspective on AI-human collaboration in this section, I now turn to examine specific collaboration models at the task level in educational settings.

Types of Human-AI collaboration on the task level

This section describes different types of teacher-AI collaboration, which have been identified in economic studies (Milanez, 2023[17]), namely full and partial replacement, complementation, and augmentation in education.

As indicated before, examples of cases in which AI replaces teachers in classrooms are rare, but examples of partial replacement are more common in classroom situations. An example of full replacement is remedial teaching, supporting students with knowledge gaps. Intelligent tutor systems can help these students practice in order to remediate these gaps without additional support from teachers (VanLehn, 2011[18]). An example of partial replacement is an adaptive learning technology that provides learners with direct feedback after solving math problems (Knoop-van Campen, Wise and Molenaar, 2021[19]). This gives learners immediate insights into the correctness of their answer, which can facilitate learning. Yet, most adaptive systems cannot provide elaborative feedback to learners explaining why they made a mistake and advising steps to resolve this error. This elaborative feedback remains the task of the teacher, and in this example, AI and the teacher execute separate elements of the feedback task. In partial replacement, mostly existing teaching practices are replaced by AI without changing or adjusting the task.

This is also the case for complementation, where AI extends teachers’ existing task execution. For example, in one of the National Education Lab AI co-creation projects, teachers asked for more advanced insights into students’ technical reading skills development over time (Radbound Universiteit, 2025[20]). Assessment is only possible every 6 months as it takes 15 minutes per child to administer the required individual test. Concurrent insights into students' reading errors based on automatic speech recognition could help teachers follow learners' reading skill development continuously and offer opportunities to attune instruction and practice materials to learners’ needs. The complementation of teachers' insights into students' development would not be possible without the additional assessment information provided by the AI application.

Another example of complementation is using AI to measure how learners plan and monitor essay writing (van der Graaf et al., 2021[21]). This application of AI allows teachers to understand the learners’ writing process, allowing for instructional practices beyond focusing on the essay. Both students and teachers can benefit from extended insight into students’ self-regulated learning activities. It helps teachers to understand existing student practices and discuss future improvements in writing and self-regulated learning processes. These examples of complementation illustrate how the additional information analysed by AI applications can create new insights that improve pedagogical and instructional practices in schools.

AI can also allow new pedagogical didactical practices that would not be possible without its use, called augmentation. An often-used example is using adaptive learning technologies and intelligent tutoring systems that select practice problems for students within their zone of proximal development. This type of differentiation at the individual and class level would not be feasible without the additional role of AI applications in classrooms. Also, AI can be used to observe how toddlers play with blocks and learn about mathematical constructs while doing so. In classroom situations, it is impossible for teachers to assess that type of detail from play behaviour, while it can be very informative of the child's development. Providing individual children with feedback while working with the blocks is feasible with an AI. These practices show how AI applications perform classroom actions, transforming pedagogical and instructional practices.

These examples show us that different types of teacher-AI collaboration are realised between the teacher and the AI. Full or partial replacement is demanded by teachers mostly for repetitive tasks, such as providing corrective feedback, simple assessment, and simple differentiation practices, as task selection. AI could potentially outperform teachers on these tasks, and the time savings created by replacement can be used for more meaningful human pedagogical and instructional practices by the teachers. In this type of teacher-AI collaboration, a clear task division depends on the functionalities that AI can perform well.

Complementation of human tasks by AI often allows for more in-depth insights into how students have performed the task, which allows teachers to extend the current pedagogical or instructional task compared to the situation before the use of AI. Here, AI adds a unique subtask to the existing task execution, widening the task scope. Again, a clear task division exists depending on the AI performance. It may not lead to time savings, but it may potentially improve quality.

Finally, in case of augmentation, the teacher-AI combination engages in the execution of a novel task, that was previously not feasible or scalable within regular classroom practices. Combined task execution makes novel practices possible. I will illustrate this with the example of curriculum adaptivity (Radbound Universiteit, 2025[22]). The AI applications can predict a class's performance and propose changes to the lesson plan to optimize the curriculum accordingly. For example, one class will need more practice on fractions, whereas another class may need more time on addition with large numbers. Each class will benefit from a different curriculum plan, which AI can predict in advance. At the same time, human teachers have unique knowledge about their class, their needs, and each child's unique requirements. This knowledge can help teachers to fine-tune and adjust the AI model predictions. In this example, the teacher and AI excel in subtasks that help refine the class curriculum. Due to the unique contributions in teacher and AI tasks, this novel task can be done in close human-AI collaboration.

Co-evolving professionalization and innovation

Research indicates that humans often misinterpret the ability of AI applications, which jeopardizes correct task allocation (Vaccaro, Almaatouq and Malone, 2024[3]). Humans are often inclined to over-rely on AI and underestimate the value of their judgments. Teachers also often view AI’s suggestions as strong guidelines without seeking and processing additional information or relying on their pedagogical instructional knowledge. Therefore, careful consideration of subtasks in teaching and the potential of AI to perform elements in these tasks is critical in designing human-AI collaboration in education. Future research endeavours should include analysis of the baseline performance of the human and AI alone, the type of task they perform, and an in-depth understanding of the division of labour between them.

In the quest to better understand teacher-AI collaboration, a new methodology will be needed (Molenaar, 2024[1]). The purpose is to create a common approach to understand human-AI collaboration at the task level and investigate the task division between teachers and AI. Existing scientific knowledge on learning and teaching can be aligned with this methodology, which may help integrate pedagogical and instruction models in designing teacher-AI collaboration. Appropriate allocation of tasks to each partner is crucial for optimal collaboration, which requires a thorough understanding of each actor's task performance. A final complicating factor is the co-evolution of human-AI collaboration over time, as humans and AI can learn from each other.

Human-AI collaboration is indeed a dynamic relationship which evolves over time. Both partners can develop in the new exchange between them. Teachers can learn from the AI, which may affect teacher practices. For example, we know that dashboard information influences teachers' feedback practices and allocation of feedback (Knoop-van Campen, Wise and Molenaar, 2021[19]). The usage of adaptive learning technologies that select personalised problems for students changes the instructional practices. Consequently, teachers often request new AI functionalities that can further induce task changes between teachers and AI.

This co-evolvement between teachers' practices and the functioning of AI is likely also to influence teachers' skills and knowledge (Cijvat et al., 2023[23]). This complex interaction may further the implementation of AI in education, in which initial replacement or complementation may lead to the transformation of teacher practices and eventually incentivize augmentation. As such, fruitful teacher-AI collaboration can lead us on a transformative path in which teachers learn to advance their pedagogical and instructional practices in interaction and collaboration with AI solutions. In this envisioned hybrid future, teacher professional development and the refinement of AI in education go hand and hand and reinforce each other in novel pathways of combined professional development and innovation.

Toward the hybrid future

These pathways that combine teacher professional development and innovation in classrooms with AI are my envisioned future. This requires a central position of teachers' needs in designing, developing, and deploying AI in education. Within the National Education Lab AI (Radbound Universiteit, 2025[24]), two programs are addressing this: our co-creation and co-implementation program (Molenaar and Sleegers, 2023[25]). Co-creation is directed at developing novel teaching practices with AI based on teachers' requirements and problem sketches. In collaboration between teachers, scientists, and developers, new AI applications are developed. The design process is focused on bridging between the teachers' pedagogical and instruction practices and the functionally of AI. Scientific knowledge on learning and teaching is focal in this development, leading to evidence-based AI solutions. These projects are human-centric and facilitate teacher complementation and augmentation with AI.

The most promising AI solutions that have proven to support human learning and teaching will be selected for our co-implementation project. Sustainable and responsible scaling of these innovations in schools and across schools is facilitated in this programme. Wider usage of AI solutions helps us further refine the connection between teacher practices and AI functionality. Acceleration pathways are trajectories to advance innovation in schools and help teachers to move forward from replacement, to complementation, and augmentation, and are referred to as human-centric application of AI in classrooms.

In combination, co-creation and co-implementation foreground the needs of teachers and learners, connect research and development towards responsible and sustainable use of AI in education. A clear focus on teacher-AI collaboration, insights into characteristics of this successful partnership, and a profound understanding of the collaboration mechanism will help us on a pathway to my envisioned future. One in which educational AI is based on teacher needs and informed by scientific knowledge, developed from an educational perspective, enabling novel educational practices that complement and augment human teachers.

These are critical steps to support the responsible future use of AI in education, in which we learn how to leverage AI to support teachers' teaching and young learners’ learning. The combination of teachers' autonomy and AI automation will be investigated to make meaningful progress in this space. In this essay, I have attempted to broaden the perspective on the effectiveness of teacher-AI combinations for learning and teaching, and an approach for successful co-creation and co-implementation of AI in education. Novel Research and Development programmes directed at designing and developing innovative teaching practices are essential to furthering our understanding of responsible and effective human-AI collaboration.

References

[14] Akata, Z. et al. (2020), “A Research Agenda for Hybrid Intelligence: Augmenting Human Intellect With Collaborative, Adaptive, Responsible, and Explainable Artificial Intelligence”, Computer, Vol. 53/8, pp. 18-28, https://doi.org/10.1109/mc.2020.2996587.

[8] Aleven, V. et al. (2023), “Towards the Future of AI-Augmented Human Tutoring in Math Learning”, in Communications in Computer and Information Science, Artificial Intelligence in Education. Posters and Late Breaking Results, Workshops and Tutorials, Industry and Innovation Tracks, Practitioners, Doctoral Consortium and Blue Sky, Springer Nature Switzerland, Cham, https://doi.org/10.1007/978-3-031-36336-8_3.

[23] Cijvat, C. et al. (2023), “Wat leraren doen: de inzet van adaptieve leermiddelen in het basisonderwijs [What teachers do: using adaptive learning tools in primary school]”, Pedagogische Studiën, Vol. 100/1, pp. 1-34, https://doi.org/10.59302/ps.v100i1.13989.

[10] Costache, O. (under review), Current State of Research on Artificial Intelligence in Teacher Education and Professional Development..

[7] Cukurova, M. (2024), “The interplay of learning, analytics and artificial intelligence in education: A vision for hybrid intelligence”, British Journal of Educational Technology, Vol. 56/2, pp. 469-488, https://doi.org/10.1111/bjet.13514.

[2] Cukurova, M., C. Kent and R. Luckin (2019), “Artificial intelligence and multimodal data in the service of human decision‐making: A case study in debate tutoring”, British Journal of Educational Technology, Vol. 50/6, pp. 3032-3046, https://doi.org/10.1111/bjet.12829.

[16] Dellermann, D. et al. (2019), “Hybrid Intelligence”, Business & Information Systems Engineering, Vol. 61/5, pp. 637-643, https://doi.org/10.1007/s12599-019-00595-2.

[19] Knoop-van Campen, C., A. Wise and I. Molenaar (2021), “The equalizing effect of teacher dashboards on feedback in K-12 classrooms”, Interactive Learning Environments, Vol. 31/6, pp. 3447-3463, https://doi.org/10.1080/10494820.2021.1931346.

[13] Lassébie, J. and G. Quintini (2022), “What skills and abilities can automation technologies replicate and what does it mean for workers?: New evidence”, OECD Social, Employment and Migration Working Papers, No. 282, OECD Publishing, Paris, https://doi.org/10.1787/646aad77-en.

[5] Ley, T. et al. (2025), “Teaching with AI: The Role of Teachers in the Hybrid Intelligent System”, in Lecture Notes in Computer Science, Two Decades of TEL. From Lessons Learnt to Challenges Ahead, Springer Nature Switzerland, Cham, https://doi.org/10.1007/978-3-032-03873-9_3.

[17] Milanez, A. (2023), “The impact of AI on the workplace: Evidence from OECD case studies of AI implementation”, OECD Social, Employment and Migration Working Papers, No. 289, OECD Publishing, Paris, https://doi.org/10.1787/2247ce58-en.

[1] Molenaar, I. (2024), Human-AI collaboration in education: The hybrid future, Radboud University, https://repository.ubn.ru.nl/handle/2066/316067 (accessed on 4 May 2026).

[11] Molenaar, I. (2022), “The concept of hybrid human-AI regulation: Exemplifying how to support young learners’ self-regulated learning”, Computers and Education: Artificial Intelligence, Vol. 3, p. 100070, https://doi.org/10.1016/j.caeai.2022.100070.

[6] Molenaar, I. (2021), Personalization of learning: Towards hybrid human-AI learning technologies, OECD Publishing, Paris, https://doi.org/10.1787/589b283f-en.

[25] Molenaar, I. and P. Sleegers (2023), Multi-stakeholder collaboration and co-creation: Towards responsible application of AI in education, OECD Publishing, Paris, https://doi.org/10.1787/c74f03de-en.

[22] Radbound Universiteit (2025), Co-implementatiepilot, https://www.ru.nl/onderzoek/onderzoeksprojecten/co-implementatiepilot (accessed on 4 May 2026).

[20] Radbound Universiteit (2025), Learning to read technically with ASR, https://www.ru.nl/en/research/research-projects/learning-to-read-technically-with-asr (accessed on 4 May 2026).

[24] Radbound Universiteit (2025), National Education Lab AI, https://www.ru.nl/en/nolai (accessed on 4 May 2026).

[12] Shneidermann, B. (2022), Human-Centered AI, Oxford University Press.

[15] Siemon, D. et al. (2017), “One for all and all for one - towards a framework for collaboration support systems”, Education and Information Technologies, Vol. 24/2, pp. 1837-1861, https://doi.org/10.1007/s10639-017-9651-9.

[4] Tammets, K. and T. Ley (2023), “Integrating AI tools in teacher professional learning: a conceptual model and illustrative case”, Frontiers in Artificial Intelligence, Vol. 6, https://doi.org/10.3389/frai.2023.1255089.

[9] Topali, P. et al. (2025), “Pedagogical considerations in the automation era: A systematic literature review of <scp>AIEd</scp> in K‐12 authentic settings”, British Educational Research Journal, https://doi.org/10.1002/berj.4200.

[3] Vaccaro, M., A. Almaatouq and T. Malone (2024), “When combinations of humans and AI are useful: A systematic review and meta-analysis”, Nature Human Behaviour, Vol. 8/12, pp. 2293-2303, https://doi.org/10.1038/s41562-024-02024-1.

[21] van der Graaf, J. et al. (2021), “Do Instrumentation Tools Capture Self-Regulated Learning?”, LAK21: 11th International Learning Analytics and Knowledge Conference, pp. 438-448, https://doi.org/10.1145/3448139.3448181.

[18] VanLehn, K. (2011), “The Relative Effectiveness of Human Tutoring, Intelligent Tutoring Systems, and Other Tutoring Systems”, Educational Psychologist, Vol. 46/4, pp. 197-221, https://doi.org/10.1080/00461520.2011.611369.