In 2026, 52% of students report that at least some of their courses provide no clear guidance on AI use. Bowen and Watson published a substantially updated second edition on December 2, 2025 to address this gap, not through prohibitions, but through a more demanding thesis. What determines whether a student develops judgment or delegates their thinking is the decision to use the tool, not the tool itself.
The essentials
- The authors present AI as a tool whose uses and effects must be evaluated against bias, ethics, and pedagogical objectives: it is the choice to delegate or not delegate a cognitive act that determines whether the student learns or circumvents learning.
- In 2026, 52% of students report that at least some of their courses lack clear rules on AI use (Lumina Foundation–Gallup 2026).
- Bowen and Watson propose a practical framework: distinguishing delegable acts (documentary research, formatting) from non-delegable acts (evaluative judgment, autonomous resolution of a new problem).
- The deeper tension is not technological but institutional: without coordinated policies, the generation trained in this interval risks consolidating a reflex of systematic outsourcing.
- The book is a working tool for teachers, not a manifesto: it delivers course design protocols tested in American universities.
Two pedagogical specialists facing an accomplished fact
José Antonio Bowen is president emeritus of Goucher College (Maryland) and one of the most cited American theorists on educational innovation. C. Edward Watson directs the Center for Teaching Excellence at the University of Georgia. Together, they published a first edition of Teaching with AI in spring 2024, several months after ChatGPT’s launch, which became the reference tool in higher education teacher training programs across America.
The second edition, published in December 2025, starts from scratch on one point. The first still attempted to map an emerging tool. The second departs from the finding that the debate on adoption is closed: AI is in classrooms, with or without institutional approval. The question has changed in nature.
The thesis: evaluate the decision, not the product
The heart of the book rests on a distinction the authors establish in the opening pages. Every pedagogical task can be broken down into cognitive acts: finding information, synthesizing arguments, identifying a contradiction, formulating a judgment. Some of these acts can be delegated to a machine without compromising the intended learning. Others cannot, because it is precisely by accomplishing them that the student builds the competency or mental disposition the course is meant to develop.
A student who asks ChatGPT to generate an essay outline before writing may have saved time. They may also have bypassed the exercise through which they would have learned to rank arguments. Bowen and Watson argue that pedagogy must reorganize itself around this dividing line. Assessment should more clearly reveal effort, process, critical thinking, and specifically human quality, and should value the quality of reasoning the student can demonstrate independent of the machine.
This thesis has an immediate practical consequence for teachers: design tasks where the choice not to delegate is visible and valued. A student who explains why they chose not to use AI on a certain part of their work demonstrates a meta-judgment that the machine cannot simulate.
Concrete tools: a pedagogy of choice
The book is first and foremost a manual. Bowen and Watson describe protocols tested in American universities. Several warrant attention.
The first is what they call the AI audit: an exercise in which students document their AI use and explain their choice. The exercise aims at critical analysis of the student’s own work process. Institutions that have adopted it report that students think more carefully when formulating a query.
The second is reorganizing assessment around cognitive acts that cannot be reproduced in a session. An unexpected oral on the content of written work, a time-limited revision, an oral justification of reasoning choices: these formats render outsourcing worthless, because the student must demonstrate an understanding the machine cannot provide at that moment.
The third is designing explicit syllabi on AI use indicating, for each task, what can be delegated and what cannot. Bowen and Watson provide templates teachers can adapt. The goal is to move beyond the ambiguity that, in its absence, benefits default outsourcing.
The authors cite experimental or quasi-experimental studies conducted in educational contexts, some reporting gains in critical thinking under certain usage conditions.
Cowen’s argument: what pedagogy owes to economics
Tyler Cowen formulated a thesis in 2024 that directly illuminates Bowen and Watson’s book. In a text published in The Free Press, he argues that AI amplifies human intellect if cognitive effort is preserved, and that societies risk competitive losses in domains where human judgment remains essential. Training critical minds thus constitutes a matter of long-term democratic competitiveness.
Bowen and Watson operationalize this thesis at the scale of a classroom. The convergence between Bowen and Watson’s pedagogical distinction and Cowen’s thesis is an interpretation of the article, not an explicit prescription from Cowen. Where Cowen poses a geopolitical diagnosis, the two pedagogues propose a concrete institutional response.
The tension between the two levels of analysis is real. Cowen notes that large enterprises may want to limit competition from cheaper, lower-quality actors. Bowen and Watson do not propose to regulate the tool, but to structure the learning environment around it. This distinction matters: their approach combines structuring learning with explicit AI use policies.
Questions the book leaves open
An honest critique of the guide lies in its blind spots. The first is unequal access to advanced pedagogical tools. The protocols described require well-trained teachers, institutions capable of investing in continuing education, and reasonably sized classes. American universities with the most resources are those most likely to adopt these methods quickly, which risks widening the gap with publicly funded institutions under budget pressure. On this point, this journal’s article on American public university budgets is a direct counterpoint: the least well-resourced institutions are also those where the institutional void around AI is deepest.
The second blind spot is large-scale assessment. Bowen and Watson cite results from pilot programs, but large-scale trials measuring long-term impact on critical thinking are still lacking.
The third is the question of international generalization. The book is written for the American university system, with its particular institutional features. More centralized systems—France, South Korea, Japan—have different levers for imposing coordinated policies, but also different rigidities. The Bowen-Watson framework is exportable in principle, less so in its modalities.
Daron Acemoglu, whose work on technology and power shows that the gains of progress do not distribute automatically, would offer a useful counterpoint here: a pedagogy of judgment that does not go hand in hand with explicit public policy on who finances teacher training will remain the privilege of the well-resourced. The book poses the right pedagogical question; it leaves the political question open.
Likely gains and losses for the 2035 generation
The real dividing line is not between institutions that permit AI and those that forbid it. It runs between those that will have structured the learning environment around the decision to delegate, and those that will have allowed this void to be filled by the reflex of outsourcing.
Several possible trajectories emerge for the coming decade. If institutions adopt coordinated policies accompanied by pedagogical devices that make student reasoning, critique, and verification observable, students could develop critical thinking structured around human decision. AI would become a tool subordinate to reasoning, as the calculator has been to mental arithmetic for forty years. The signals to watch here are simple: the share of institutions with a formalized and enforced AI use policy, measured annually, and the evolution of results in standardized critical thinking assessments in post-2023 cohorts.
The other trajectory is less comfortable. In a 2024 British survey of 1,250 students, 12% found institutional AI policy unclear, while 63% found it clear; the consequences of this ambiguity on delegation habits are not established. A lasting consolidation of this reflex risks affecting the quality of judgment assessments. This risk is not theoretical: massive AI adoption in organizations is already creating demand for workers capable of evaluating model outputs rather than producing them, a relevant component of critical thinking and AI literacy.
Essential collective needs remain in both scenarios: assessment methods capable of distinguishing student reasoning from generated content, and teachers trained to design tasks where the choice not to delegate is visible. The proposed path combines institutional and course policies, curricular design, and concrete assessment redesign: defining essential competencies, usage policies, and AI-adapted assessments. Bowen and Watson’s book supplies the pedagogical substance. It remains to mobilize the political will necessary to adopt it at scale.
Interest of the book
Teaching with AI first addresses higher education teachers. But anyone questioning what generative AI changes in the formation of minds will find in it a rigor that usual opinion pieces on the subject do not offer. The book avoids two symmetrical pitfalls: moral panic at widespread cheating, and naive enthusiasm for a tool that would make learning faster. It poses a harder question—what does it mean to learn, and what does delegating to the machine prevent from being learned—and answers it with tested protocols, not intuitions.
For a French-speaking reader, the book offers another interest: it documents what is happening in American universities at the precise moment European systems are beginning to ask themselves the same questions without yet having the same field experience. Solutions are not directly transposable, but the problems are identical.
Bibliographic information
Teaching with AI: A Practical Guide to a New Era of Human Learning José Antonio Bowen, C. Edward Watson Johns Hopkins University Press 2nd edition, December 2025
Sources
- Teaching with AI, Johns Hopkins University Press
- Tyler Cowen, A Doomsday Scenario for American AI, The Free Press, https://www.thefp.com/p/tyler-cowen-ai-doomsday-scenario-american
- Stanford Tech & Education Guide 2026, Stanford University (no link: institutional page not stabilized)
