Research Article
From Technological Adaptation to Professional Redefinition: Transformation of University Teachers' Roles and Educational Empowerment in the Context of Human–AI Synergy
Synthesis: Yang and colleagues (2026) argue that the transformation of university teaching under human–AI synergy is not the addition of a tool skill to an existing professional repertoire. It is a shift from knowledge-transmission professionalism to what they call educational professionalism in the digital intelligence era: the capacity to hold the formation of the person as the governing criterion, to translate talent goals into authentic tasks, to allocate responsibility among teacher, student and intelligent system, to improve teaching from evidence of learning, and to judge the boundaries, risks and ethical responsibilities of technology use. They describe five shifts in the surrounding ecosystem — talent development toward future-oriented competence, knowledge provision toward multi-actor collaboration, teaching toward the organization of deep learning, knowledge production toward human–AI co-intelligence, and assessment toward process evidence and comprehensive judgment — and then recast the teacher as value leader, learning designer, human–AI collaboration organizer, and teaching researcher. The unit of reform, on this account, is the institution.
No primary data: this is a normative commentary assembled from literature and from policy documents of the authors' own university and of the Chinese Ministry of Education. The trials it cites were conducted by other researchers in other settings, and every figure below is attributed to them rather than reported here.
Key Findings
- The unit of change is professionalism, not a skill set. The proposition is stated early: the key is not teaching faculty who never used AI how to use it. If the knowledge-authority model survives intact and a tool is bolted onto it, the teacher still stands in front of students as the source and validator of content — the position GenAI has undermined. Their alternative defines digital-intelligence professionalism by five capacities: holding the formation of the person as the governing criterion, translating talent goals into authentic tasks, allocating responsibility among teachers, students and systems, improving teaching from evidence, and judging boundaries and risk. That is a claim about identity, and it implies development work longer and less modular than tool training.
- Five ecosystem shifts each remove a function the teacher performed by default. Talent development moves from knowledge mastery to future-oriented competence, which at Tianjin University became a six-element talent profile and a plan that shifts the objective from solving known problems to exploring the unknown. Knowledge provision moves to a ternary teacher–student–AI structure with explicitly unequal responsibility: teachers set goals and assure quality, students inquire and verify, systems generate resources and give repetitive feedback. Teaching moves from presenting content to organizing deep learning. Knowledge production moves to human–AI co-intelligence, with frontier questions reaching the curriculum faster. Assessment moves from outcome determination to process evidence.
- Whether AI helps depends on instructional design, and the cited evidence runs both ways. They report a randomized crossover trial at Harvard with 194 undergraduates in a physics course, where an AI tutor built on learning-science principles produced a median learning gain more than twice that of the classroom human tutor, at a median 49 minutes against 60 (Kestin et al., 2025). They set against it a field experiment with nearly 1,000 secondary-school mathematics students, where an unguarded assistant raised practice performance by 48% but left students 17% lower than controls on an unassisted examination, an effect that largely disappeared once teachers added designed prompts, error guidance and scaffolds (Bastani et al., 2025). Their reading is the load-bearing one: access is not the variable, design is.
- AI literacy is a judgment capacity, measured by what a teacher decides rather than by what a teacher can operate. Literacy, on their account, is not how many tools a teacher masters or how many resources a teacher generates, but whether the teacher can choose technology by educational purpose, verify output, allocate responsibility and control risk. That definition connects the role argument to the governance documents they cite: the 2026 action plan and code of conduct at Tianjin University specify encouraged and prohibited uses and require users to identify the tool, the scope and the intended purpose. Disclosure becomes a design requirement of the institution rather than a matter of individual conscience.
- Four roles, each a demotion of transmission and a promotion of judgment. The value leader holds ethics and conduct as the primary criterion and integrates value formation into objectives, content and feedback, on the argument that the more an era prioritizes efficiency, the more education must preserve human warmth. The learning designer moves from explaining content clearly to making learning happen, through differentiated pathways and higher-order questioning. The collaboration organizer designs the ternary structure rather than occupying its center. The teaching researcher turns implementation problems — how students formulate questions, how they verify outputs — into inquiry. Authority is exercised over design, not over information.
- The support system has to change with the role, or the redefinition is exhortation. The fourth section is the paper's most institutional and least evidenced. It proposes top-level design embedding human–AI collaboration, digital ethics and interdisciplinarity into a competency framework; an infrastructure that collects teaching-behavior data across academic affairs, training and research systems, diagnoses each teacher's profile; a four-tier training system from new through core, distinguished and leading teachers; and evaluation reform creating teaching-oriented posts and promotion channels so teaching carries the same weight as research. The last is what the rest depends on, and the authors know it: a role that cannot be rewarded will not be adopted.
What this means for practice
- Write the target capability as observable performance. "Uses AI well" cannot be developed or evaluated. A department can specify what it looks like instead: selects a tool against a stated learning problem, verifies an output against a source, assigns and explains responsibility in a task, and states what remains under human control.
- Make the division of responsibility explicit in every AI-supported task. State what the teacher decides, what the student must do and be able to defend, and what the system is permitted to produce. Students who are told only that AI is allowed will distribute responsibility themselves, usually toward the system.
- Assess the process the task was built to develop. Question formation, source selection, prompt iteration, comparison of alternatives, verification, oral explanation and reflective revision are all observable. A submission that cannot be defended orally is evidence about the artifact, not about the learning.
- Sequence training by career stage rather than by tool. New faculty need ethical standards and safe handling; experienced faculty need course-based inquiry into AI-supported learning; senior faculty need to lead standards and interdisciplinary work. The progression is in the object of inquiry, not in the software.
- Fix the reward structure before announcing the role. If teaching counts less than research in promotion, the redefinition will not survive the first review cycle. Teaching-oriented posts and equivalent weighting are the precondition, not the follow-up.
Limitations
- There is no method. The abstract claims literature analysis and empirical evidence, but the paper reports no search strategy, no inclusion criteria, no coding procedure and no data of its own. It is a position paper with citations.
- The empirical foundation is two studies from other populations. One is a single randomized trial with undergraduates in one physics course; the other is a field experiment with secondary-school students learning mathematics. Neither concerns university faculty, and the paper does not ask whether such effects transfer to the teaching of law or design.
- The institutional examples are the authors' own employer, described rather than evaluated. The talent plan, the action plan, the code of conduct and the platform are presented from policy documents, with no implementation data, no faculty response, no learning outcome and no indication of cost.
- "Educational professionalism" is defined by enumeration and cannot be measured. Five capacities are listed with no instrument, no threshold and no evidence that teachers who display them produce better learning. A construct that cannot be observed cannot show whether a training program worked.
- The data infrastructure is proposed without a privacy discussion. A system that assembles a competency profile for each teacher from academic affairs, training and research records is a surveillance arrangement as well as a development service. Consent, error, appeal and use in personnel decisions are not addressed.
- The argument is normative and universal in scope. It does not consider institutions without the data infrastructure, design staff or promotion flexibility the four paths assume, nor does it say what a faculty member who judges the redefinition wrong should do.
Citation
Yang, X., Xia, S., Liu, X., Zhang, Z., Huang, L., & Wang, L. (2026). From Technological Adaptation to Professional Redefinition: Transformation of University Teachers' Roles and Educational Empowerment in the Context of Human–AI Synergy. Frontiers of Digital Education, 4(1), article 4.