Leading teachers' perspective on teacher-AI collaboration in education

Abstract

Moving beyond the direct support all alone by a human teacher or an Artificial Intelligence (AI) system, optimizing the complementary strengths of the two has aroused great expectations and educational innovation potential. Yet, the conceptual guidance of how best to structure and implement teacher-AI collaboration (TAC) while ensuring teachers' instructional roles to support students learning remains limited. This study, therefore, aims what (1) curriculum, (2) teacher-AI interaction, (3) learning environment would be required as well as how TAC would evolve by reflecting teachers' views. Through in-depth interviews with 20 Chinese leading teachers in AI in Education (AIED), the study found that teachers aimed to improve students' subject-matter knowledge and build their capacity as the desired goals for TAC and these can be carried out by data-driven problem-based learning and case-based reasoning in tandem with growth-focused and reflective assessment. While teachers highlighted that developing teachers' data literacy and collegiality with AI are essential, they expected AI to be equipped with Technological Pedagogical and Content Knowledge (TPACK) knowledge and conflict resolution skills. In addition, teachers expressed that Internet of Things (IoT)-based classrooms, systematic AIED curriculum, school-based continuing professional development, research-practice-policy partnerships as well as creating a continuous learning and AI-ready culture are significant. Furthermore, teachers envision TAC would develop into three stages: (1) teachers as passive AI recipients, (2) teachers as active AI users (3) teachers-AI as constructive partners. These findings build a more holistic and in-depth understanding of the AIED and offer implications for the educational AI design and teachers' education.

Appendix 1: Summary of emergent themes

Category		Themes	Sub-themes	Exemplary quotes
Curriculum (RQ1)	Learning goal	T1. Improving subject-matter knowledge		We use an AI-based English platform that aims to promote students’ reading comprehension by providing texts at different difficulty levels. (P7)
		T2. Building students capacity	Cognitive capacity	For our students to work with AI in the digital society, CT is must-have competency for them to present problems in a way that AI and other tools can understand and help solve them. (S3)
			Social capacity	Students interrogate multiple sources of data to discover social issues they were not conscious of such as racism, sexism, and structural poverty. The class gradually debates about what sources of data must be varied and analyzed and learn how to conduct and complete an ethical audit of real-world scenarios working with data and AI. (S2)
			Emotional capacity	Students talk about their academic pressure, friends and family relationships and career plan ambiguity, and so on with the AI-based Chabot. Reflecting on the analysis of their text-based conversation, the school counsellor teacher and homeroom teacher work together to identify critical situations that influence individual students' moods and behavior and instruct them on how to handle those challenges by correcting their thinking and behavioural reactions. (S10)
	Content	T3. Data-driven problem-based learning		Students can be engaged in data-driven inquiry as they explore a rich data set and observe patterns, ask questions suggested by the data, and pursue investigations about underlying phenomena in collaboration with their peers and teachers. (S7)
		T4. Case-based reasoning		As creativity is just connecting things, students can better learn things by reflecting on previous cases, experiences, and knowledge that are naturally brought to bear in interpreting new situations. In fact, searching for cases and contents is one of the major tasks that sucks our time. I Hope AI can build up the case library, streamline appropriate cases and sources into different classroom instruction, suggest students different points, facilitate their inferences, and so on. (S4)
	Assessment	T6. Growth-focused assessment		AI's accumulated students' data over time would enable teachers to recognize the evolution of students' learning progress and be able to provide the necessary suggestions for further improvement.
		T7. Reflective assessment		While AI interacts with students, it could guide students to reflect on mistakes to avoid repeating them, encourage them to consider and comment on their learning experiences- not only what they learned but also how they did it, and build their own understanding and viewpoint. (P8)
Teacher-AI Interaction (RQ2)	Cognitive interaction	T8. Teachers’ data literacy		Ability to extract information from subject-specific data for further processing using AI tools, conduct efficient data sanitization and mitigate unfairness in dataset using AI methods, implement concrete AI learning procedures, raise questions about AI's prediction and analysis and explain why results contain errors and questions. (P5)
		T9. Intelligent-TPACK		The true power of AI depends on its pedagogical knowledge to understand teachers' behavior and decisions and facilitate teachers' practice simultaneously. (P1)
	Social interaction	T10. Developing collegiality with AI		Although AI cannot be exactly the same as a human colleague, it could at least be treated as a machine teaching partner to expand TAC scope, delegating more tasks to AI and reviewing and reflecting on each others' teaching practices. In turn, I can train AI to deliver more meaningful instruction to students and AI can guide me to be a better version of myself. (S8)
		T11. Conflict resolution skills with teachers		Classroom instruction is not simply about realizing learning activities and delivering learning content but also about managing classroom conflict. AI then needs to well understand the different natures of conflict, for instance learning through a crisis can be beneficial, learn about and reflect upon the various social relationships within students, and monitor how these dynamics play out during classroom instruction. (P10)
	Artifact-mediated interaction	T12. Edge computing		All data processing better occur on the device itself for teachers to take input from students and offer students progressive feedback that can better their cognitive responses and thought processes. (S1)
		T13. Intuitive dashboard		Such massive, dynamic, and ambiguous data need to be synthesized using visual representations for teachers to explore and understand large amounts of information at once and gain insights that directly support instruction planning. (S4)
Environ-ment (RQ3)	Learning space	T14. IOT based cloud integrated classroom		More internet-connected devices, such as laptops, tablet PCs, and whiteboards, paired with data analytics technology, can help teachers and AI monitor students' learning engagement and process during classwork and testing, and ultimately provide more agile and personalized instruction. (P2)
	Institution	T15. Systematic AIED curriculum		Differences in the design and deployment of AI education across the country make it difficult to consistently define AI education. In particular, I think it is an overly difficult curriculum for young children, especially when students require significant background knowledge to understand algorithms powered by deep learning. If technical components of AI like programming are highlighted, it's limiting areas and scope of TAC as well as student-AI collaboration in the learning domain. (P9)
		T16. School-based Continuing Professional Development		Although we regularly attend the training sessions offered by the provincial department of education and even the university in the region, it's quite difficult to contextualize their experiences in the school. It should better be delivered in the school environment so that AIED is aligned with the needs of teachers of different subjects and with school-wide goals. (S5)
		T17. Research-Practice-Policy Partnerships		We need to move beyond understanding AI adoptions in schools narrowly as using existing AI tools matters of the individual teacher. Rather, considerations of ethical interaction design between teacher and AI, student, and AI, or even teacher-student-AI, curriculum design when AI is embedded in the classroom, and classroom design are considered as they relate to redressing the opportunities and harms associated with AI. Thereby, we need to maximize cross-disciplinary expertise among government, industry, and school, co-invest their different resources such as money, technology, information, knowledge, facility and etc., and make them interact in the decision-making process to leverage broader educational benefits and accelerate educational-oriented AI technological solution. (S9)
	Culture	T18. Continuous learning culture		Teachers' professional development is usually evaluated based on the number of AI-related programs attended with little focus on the transfer of such learning to the actual teaching practice. There is a need to create a fundamental change in school culture to lead every teacher to learn new technology and AI and apply innovative pedagogy with technology, and they also need to feel compelled to share their knowledge with others. (S6)
		T19. AI-ready culture		School first needs to become AI-ready school which requires a fundamental transformation in how teachers do things, relate to each other, what skills we have, and what processes and principles guide our behaviors. In particular, not only the school highlights a wide new range of AI-related skills and competencies, but everyone in the school needs to be in a position to obtain insight into practical applications across many tasks and activities in the school. (P4)
Co-evolution (RQ4)		T20. Teachers as passive AI recipients		Teachers begin to use AI as one of the existing school technology tools to deliver curriculum content to students, mainly using it to support the delivery of a lecture, and use it for a simple learning activity for their easy control and management. (P3)
		T21. Teachers as active AI users		As teachers have a greater familiarity with the use of AI tools and have a more conceptual understanding of AI, they will incorporate AI tools as an integral part of the instruction. Teachers design more effective AI-mediated learning experiences by relating AI's affordances with student-centered learning pedagogy. (P6)
		T22. Teacher-AI as constructive partners		Assuming that AI technology is much more advanced, teachers and AI will somehow work as partners not only to develop students' intellectual capacity, but also to support them emotionally, socially, and ethically who live lives of meaning and purpose. To do so, teachers and AI will work closely to monitor and evaluate students learning process with their own strengths and expertise across on/offline classroom instruction. (S1)