A principled way to think about AI in education: guidance for educators and policy makers based on goals, models of human learning, and use of technologies

Noah D. Finkelstein

REVIEW 2 major objections 5 minor 9 references

A principled way to think about AI in education: guidance for educators and policy makers based on goals, models of human learning, and use of technologies

T0 review · 2 major / 5 minor · reviewed 2026-08-04 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper claims that explicit principles built from educational goals, a socio-cultural account of learning, and a functional definition of technology can keep generative AI from displacing the human core of higher education.

desk verdict A clear and honest principled framework for AI in higher ed: more practical synthesis than new research, with its key socio-cultural premise asserted rather than proven, but that is acceptable for a position piece. read the letter →

arxiv 2510.01467 v2 pith:3TZ7ZTT6 submitted 2025-10-01 cs.CY physics.ed-ph

Noah D. Finkelstein This is my paper · ORCID

classification cs.CYphysics.ed-ph

keywords generativeAIineducationhigherpolicylearningsciencessocio-culturaltheoryofeducationaltechnologyguidingprincipleshuman-centeredteachingand

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Finkelstein argues that educators and policymakers do not have to choose between embracing generative AI wholesale or banning it: a set of explicit principles can connect large goals for education to day-to-day choices about tools. The paper builds the framework from three definitions—what education is for (developing individuals, sustaining society, preparing a workforce), what learning is (socializing humans into cultural systems through tool-mediated interaction), and what technology is (human-constructed tools that reorganize how people interact). On that base it grounds a series of principles (P0–P2.5) for deciding which parts of teaching and learning must stay human-led and what can be outsourced to AI. If the framework is right, it offers a stable, adaptable vocabulary for evaluating AI tools as they change, and a way to make sure technology augments rather than displaces human capacities.

What carries the argument

The framework itself is the central object: a chain of three working definitions (educational goals; learning as socialization into cultural systems; technology as human-constructed tools that reorganize interaction) that generates a set of guidance principles, numbered P0 through P2.5, for educators, learners, and administrators. The load-bearing principle is P1/P2: in education, decide what domains of work are 'necessary or essential' for humans to lead and direct, and what is 'feasible and safe' to outsource to technology. This allocation rule, applied with the definition of learning, is what keeps AI augmenting rather than displacing human capacities.

What would settle it

A controlled study in a real college course where one group learns exclusively from an AI tutor with no human instructor and another from a human-led class covering the same content, measuring not only test scores but identity, motivation, community belonging, and civic engagement; if the AI-only group matches or exceeds the human-taught group on all dimensions, the paper's central allocation principle would be contradicted.

Watch

Extended reading notes

Core claim

The central claim is that the gap between AI's promise and peril and ground-level implementation can be bridged by articulating principles grounded in socio-cultural learning theory and a functional definition of technology. The framework's core move is to ask first what education is for, then what learning is, then what technology is, and only then to allocate responsibilities: educators keep human-led the work of setting goals, curating information and knowledge, designing activities, assessing, certifying, and building community; students keep human-led the practices of attending to goals, practicing skills, synthesizing across courses, collaborating, and evaluating. AI tools are to be tr

Load-bearing premise

The framework rests on the socio-cultural claim that learning is socializing humans into cultural systems, so that human-led interaction is essential to higher-order educational goals; if AI-mediated instruction alone could develop those goals as well as human-mediated instruction, the principles for keeping tasks human-led lose their basis. The paper also assumes, for the moment, that higher education's institutional value and relevance stand.

Editorial extensions

If this is right

Institutions gain a concrete audit tool: ask of any AI deployment which educational goal it serves and whether it removes human-led practice that students or educators need.
Educators have a principled rationale for keeping goal-setting, activity design, assessment, certification, and community-building human-led even when AI can perform them.
AI tutoring for basic skills is legitimate as a time-saving support, but the framework warns it can worsen inequities if it replaces caring human interaction for underprepared students.
Assessment can evolve toward fine-grained records of what students can do (micro-credentials, badges, portfolios) as long as certification retains human professional judgment.
Administrators are responsible for resourcing adoption—training, time, and workload relief—rather than simply providing access to AI tools.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A direct testable extension: any proposed AI deployment can be classified by which principle it satisfies, and deployments that remove student practice from a goal-relevant skill should be redesigned regardless of their efficiency.
The socio-cultural premise predicts that purely AI-mediated instruction will plateau at skill acquisition and miss identity, motivation, and community outcomes; longitudinal comparisons of AI-only versus human-plus-AI courses could test this.
The framework's logic implies that the instructor's role shifts toward curation, design, and judgment—so workload should shift away from clerical tasks, a point administrators can use in planning.
The principles also suggest a governance model: students co-author class AI policies, which may increase buy-in and align performance measures with learning.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

This is a position essay, not a research result. What's actually new is the explicit numbered framework—P0 through P2.5—that gives educators and administrators a shared language for connecting broad goals to concrete decisions about AI. The packaging is original, and the principles are specific enough to be useful in department meetings or faculty development sessions. The writing is plain and honest, and it doesn't oversell what it offers.

The paper does several things well. It grounds the framework in established learning science (Vygotsky, Cole, the 2025 NAS report) without being preachy. The scenarios—like students training AIs or TA shortages—help show how the principles work in practice. It also flags its own assumptions, including the explicitly temporary assumption that higher education is worth preserving. That is more upfront than most essays of this type.

The soft spots are real but not disqualifying. The socio-cultural premise—that learning is about socializing humans into cultural systems—is asserted from scholarship, not tested. The stress-test worry about Kestin et al. (2025) doesn't land as hard as it initially sounds, because the paper's opening question explicitly acknowledges that AI can deliver instruction, and the argument is that universities do more than deliver content. Still, the framework never specifies when human mediation is actually necessary versus merely preferable. That is a genuine gap, and it leaves the central recommendations underdetermined in exactly the situations the paper targets.

There are two other limitations worth noting. The paper doesn't engage with other AI-in-education frameworks, so it doesn't tell you how it differs from or improves on existing checklists—which means the novelty is packaging rather than substance. And there is no empirical validation, though for an essay meant to guide practice, that is not a fatal flaw.

Who gets value from this? Practitioners: faculty, center-for-teaching staff, deans, and policymakers who want a principled starting point for discussing AI. It won't give you new evidence or a rigorous causal claim, but it will give you a coherent vocabulary. I'd want to see it get serious peer review from someone who knows the learning sciences and the AI-in-education landscape, because the framework deserves to be refined by conversation. It's a solid contribution to the discourse, not a breakthrough.

Recommendation: engage with it for what it is—a thoughtful, well-argued position piece—and don't demand empirical proof it doesn't claim to offer.

Referee Report

2 major / 5 minor

Summary. This essay argues that higher education's response to generative AI should be organized around explicit principles derived from learning sciences rather than only around promises/perils or implementation tips. It proposes Principle 0 (make educational goals explicit), Principle 1 (keep certain educator activities human-led, with sub-principles 1.1–1.5 on goals, curation, design, assessment, and improvement), and Principle 2 (keep certain student activities human-led, with sub-principles 2.1–2.5 on learner goals, practice, synthesis, collaboration, and evaluation). The paper illustrates the framework with scenarios and returns to three opening questions about institutional survival, student motivation, and changing forms of thought. The central claim is that this principled framework bridges broad educational goals to practical implementation choices.

Significance. If the framework is accepted, it gives educators and policymakers a useful shared vocabulary for discussing AI in higher education and an antidote to purely techno-optimistic or purely prohibitive responses. Its strengths include a clear value stance, integration of established learning-science scholarship (Vygotsky, Cole, NAS 2025, Reich), and an attempt to make its assumptions explicit. It does not present new empirical evidence or machine-checked results; its contribution is a synthesis and a set of conditional recommendations. The main risk is that the framework's most load-bearing premise—that social, human-led interaction is necessary for the learning outcomes higher education values—is asserted rather than tested. Because this premise drives Principles 1 and 2 and the concluding defense of universities, the practical guidance is only as secure as that premise. The paper does, however, produce falsifiable conditional statements (e.g., 'if the goal is X, then activity Y should remain human-led') that could be examined in future studies.

major comments (2)

[What is learning?; Principle 1.3; Addressing the opening questions] The framework rests on the socio-cultural claim that 'Learning is about socializing humans into cultural systems' (Vygotsky 1978 / Cole 1996). This premise is used to justify Principles 1 and 2, which direct educators and students to retain certain activities as human-led. However, no boundary condition or evidence is given for when human mediation is necessary for the development of higher-order functions. The opening cites Kestin et al. (2025), where AI tutoring outperformed in-class active learning—the very kind of human-led active engagement that Principle 1.3, citing NAS 2025, prioritizes. That result does not decisively refute the socio-cultural premise, but it does show that for some skills, AI-mediated instruction can achieve strong outcomes without the human-led social context. The manuscript should either specify the type of learning outcome (e.g., identity formation, collabora
[Why a Principled Approach; Principle 0; Principle 1] The paper's central promise is that a set of principles can 'bridge the gap' between large-scale goals and implementation. Yet the principles are phrased as open-ended considerations ('Consider how...', 'identify which goals...', 'determine what practices...'), and terms such as 'necessary or essential' (Principle 1) and 'feasible and safe' (Principle 1) are not defined. Principle 0 acknowledges that goals can be 'multifaceted, and potentially contradictory,' but no procedure is offered for weighing or prioritizing those goals when they conflict. As a result, the framework is closer to a checklist of value-laden questions than to a decision procedure that could tell an educator or policy maker what to do in a specific case. This does not invalidate the essay, but the claim that the principles are a 'bridge' should be tempered; otherwise, the reader cannot determine whether two people fol

minor comments (5)

[What is learning?] The text 'Vygotsky1978 / Cole 1996' is missing a space. Minor typo.
[Supporting Learners' Roles] Principle numbering is inconsistent: 'Principle: 2.4' appears where other sub-principles are labeled 'Principle 2.1' etc. Standardize.
[Why a Principled Approach] The phrase 'If we do not proactively and engage' is grammatically incomplete. Also, 'We have grand agency and capacities' is informal; consider aligning tone with the rest of the essay.
[Addressing the opening questions] The conclusion that universities remain necessary is conditional on the goals 'advancing the lives of individuals and building a broader society.' This conditionality is acknowledged earlier ('assume their value and relevance (for the moment)'), but the closing reads as categorical. It would help to restate that the argument is conditional on this value commitment so that a reader who does not share the premise is not misled.
[Overall] The paper uses several non-peer-reviewed sources (e.g., New York Times articles, arXiv preprints) for current-events claims. This is acceptable for a fast-moving topic, but flagging which claims are established research versus journalistic reporting would improve rigor.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the framework is an explicit conditional argument from stated learning-science premises; the single self-citation is illustrative only.

full rationale

This paper is a position/framework essay rather than a derivation with fitted parameters or empirical predictions. Its argument chain runs from explicit goals (Principle 0; 'Why educate?'), to a socio-cultural definition of learning ('Learning is about socializing humans into cultural systems' Vygotsky/Cole), to a functional definition of technology ('Technologies are human constructed tools, both material and intellectual, that reorganize how humans interact'), to the enumerated principles P1-P2.5. No equation or fitted parameter is used, and no quantity is predicted from another. The central recommendations are conditional on explicitly stated premises—e.g., the essay says it will 'assume their value and relevance (for the moment)' regarding institutions of higher education—so if the socio-cultural premise is challenged, the practical guidance is weakened, but that is a correctness/evidence risk, not circularity. The only in-text self-citation (Ben-Zion et al. 2024) appears in one scenario as an example of students designing physics simulations; it is not used as evidence for the framework itself. The load-bearing scholarly support comes from external citations (Vygotsky; Cole; NAS 2025; Farrell et al.; Kestin et al. 2025). Thus no circular step is present; the score of 2 reflects only the presence of a minor, non-load-bearing self-citation, not derivation-by-construction.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

No free parameters (no quantitative model). Five explicit domain/ad hoc assumptions carry the framework; all are acknowledged, but none is empirically validated or compared with competing frameworks.

assumptions (5)

domain assumption Learning is socializing humans into cultural systems through mediated interaction with people and tools (Vygotsky 1978; Cole 1996).
Basis for Principles 1 and 2, which allocate human-led vs. outsourced roles; asserted from scholarship, not tested in the paper.
domain assumption Technologies are human-constructed tools, material and intellectual, that reorganize how humans interact (Cole 1996; Farrell 2025).
Definitional frame used to treat generative AI as a value-laden tool with predispositions rather than neutral.
domain assumption Education serves three broad goals: individual development, societal infrastructure, and workforce development, in that priority order.
Normative ordering explicitly attributed to the author; drives Principle 0 and all downstream role assignments.
domain assumption Institutions of higher education have value and relevance; the essay assumes this 'for the moment' (Why educate?).
If higher education's institutional form is not assumed necessary, the framework's application scenarios lose their target.
ad hoc to paper An educator's core activities can be partitioned into objectives, content, habits of mind, motivation/community, curation, design, assessment, and improvement.
This taxonomy organizes the sub-principles but is not derived from an established theory or validated against alternatives.

how reviews work

0 comments

Cite this review

Pith. "Pith review of A principled way to think about AI in education: guidance for educators and policy makers based on goals, models of human learning, and use of technologies." pith.science (2026). https://pith.science/paper/3TZ7ZTT6

@misc{pith2026251001467,
  author       = {Pith},
  title        = {Pith review of: A principled way to think about AI in education: guidance for educators and policy makers based on goals, models of human learning, and use of technologies},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3TZ7ZTT6}},
  note         = {Machine review of arXiv:2510.01467}
}

read the original abstract

The rapid emergence of generative artificial intelligence (AI) and related technologies has the potential to dramatically influence higher education, raising questions about the roles of institutions, educators, and students in a technology-rich future. While existing discourse often emphasizes either the promise and peril of AI or its immediate implementation, this paper advances a third path: a principled framework for guiding the use of AI in teaching and learning. Drawing on decades of scholarship in the learning sciences and uses of technology in education, I articulate a set of principles that connect broad educational goals to actionable practices. These principles clarify the respective roles of educators, learners, and technologies in shaping curricula, designing instruction, assessing learning, and cultivating community. The piece illustrates how a principled approach enables higher education to harness new tools while preserving its fundamental mission: advancing meaningful learning, supporting democratic societies, and preparing students for dynamic futures. Ultimately, this framework seeks to ensure that AI augments rather than displaces human capacities, aligning technology use with enduring educational values and goals. It is meant as a practical guide for anyone engaging in the use of new technology tools in their educational practices, or those people setting educational policies in the modern era. A new preamble (Jul 2026) and modest updates throughout the paper contextualize it since initial sharing (Jul 2025). The good news is that the arguments and principles of action posted originally appear to still be relevant one year later

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

9 extracted references · 1 canonical work pages

[1]

A principled way to think about AI in education: guidance for action based on goals, models of human learning, and use of technologies. Noah Finkelstein University of Colorado, Boulder Draft 7/30/25 If generative AI and associated new technological tools can deliver instruction to our undergraduates in a just-in-time and student-specific manner (Kestin, e...

2025
[3]

Learning is the mechanism by which people develop [Vygotsky 1978]

Developing individuals. Learning is the mechanism by which people develop [Vygotsky 1978]. The development of higher order cognitive functions not only supports basic skills for engaging in society (language and mathematical literacies, reasoning, argumentation, communication, socialization, empathy, agency, identities, and more), but also makes individua...

1978
[4]

physics questions

Workforce development. The most common, modern framing of the purpose of education is to get a job, or more appropriately, to support a career. Arguments around workforce development vary from specific skills training and certification to gaining broader skill sets and supporting the capacities for job and place-based learning. Motivations also vary from ...

2009
[7]

Washington, DC: The National Academies Press

Transforming Undergraduate STEM Education: Supporting Equitable and Effective Teaching. Washington, DC: The National Academies Press. https://doi.org/10.17226/28268. O’Rourke, M. (2025). I Teach Creative Writing. This Is What A.I. Is Doing to Students, NYTimes, Jul 18,

work page doi:10.17226/28268 2025
[8]

Reich, J. (2020). Failure to disrupt: Why technology alone can’t transform education. Harvard University Press. Singer, N. (2025), Welcome to Campus. Here’s Your ChatGPT., New York Times, 7 Jun

2020
[9]

Stephenson, N. (1995). The Diamond Age: Or, A Young Lady's Illustrated Primer, Bantam Books. Steincke, K.K. (1948). Farvel Og Tak: Minder Og Meninger (Farvel Og tak: Ogsaa en Tilvaerelse IV (1935-1939)), Quote Page 227, Forlaget Fremad, København. (Publisher Fremad, Copenhagen, Denmark) Vygotsky, L. S. (1978). Mind in society: The development of higher ps...

1995
[191]

https://www.carnegielearning.com/ Cole, M

Carnegie Learning (2025). https://www.carnegielearning.com/ Cole, M. (1998). Cultural psychology: A once and future discipline. Harvard University Press. Dewey, John (1916). Democracy and Education: An Introduction to the Philosophy of Education. New York: Macmillan Farrell, H., Gopnik, A., Shalizi, C., & Evans, J. (2025a). Large AI models are cultural an...

arXiv 2025
[2024]

and (2) ways to implement these technologies in our learning environments (Mollick, 2025a, 2025b) here I seek to promote a third way - a foundational approach to the above – principles to help us think about the productive uses of new technologies in education. These principles build on decades of scholarship of teaching and learning and of technology use...

1948

Show all 9 references

[2025]

which go back to the early days of the typing tutor. Arguably these approaches can support more advanced goals in a course (1.1., 2.1), utilizing our limited class time on practices that support these goals (1.3,1.4, 2.2, 2.3) and even afford the time to engage with students o...

2020 arXiv

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.

[1] [1]

A principled way to think about AI in education: guidance for action based on goals, models of human learning, and use of technologies. Noah Finkelstein University of Colorado, Boulder Draft 7/30/25 If generative AI and associated new technological tools can deliver instruction to our undergraduates in a just-in-time and student-specific manner (Kestin, e...

2025

[2] [3]

Learning is the mechanism by which people develop [Vygotsky 1978]

Developing individuals. Learning is the mechanism by which people develop [Vygotsky 1978]. The development of higher order cognitive functions not only supports basic skills for engaging in society (language and mathematical literacies, reasoning, argumentation, communication, socialization, empathy, agency, identities, and more), but also makes individua...

1978

[3] [4]

physics questions

Workforce development. The most common, modern framing of the purpose of education is to get a job, or more appropriately, to support a career. Arguments around workforce development vary from specific skills training and certification to gaining broader skill sets and supporting the capacities for job and place-based learning. Motivations also vary from ...

2009

[4] [7]

Washington, DC: The National Academies Press

Transforming Undergraduate STEM Education: Supporting Equitable and Effective Teaching. Washington, DC: The National Academies Press. https://doi.org/10.17226/28268. O’Rourke, M. (2025). I Teach Creative Writing. This Is What A.I. Is Doing to Students, NYTimes, Jul 18,

work page doi:10.17226/28268 2025

[5] [8]

Reich, J. (2020). Failure to disrupt: Why technology alone can’t transform education. Harvard University Press. Singer, N. (2025), Welcome to Campus. Here’s Your ChatGPT., New York Times, 7 Jun

2020

[6] [9]

Stephenson, N. (1995). The Diamond Age: Or, A Young Lady's Illustrated Primer, Bantam Books. Steincke, K.K. (1948). Farvel Og Tak: Minder Og Meninger (Farvel Og tak: Ogsaa en Tilvaerelse IV (1935-1939)), Quote Page 227, Forlaget Fremad, København. (Publisher Fremad, Copenhagen, Denmark) Vygotsky, L. S. (1978). Mind in society: The development of higher ps...

1995

[7] [191]

https://www.carnegielearning.com/ Cole, M

Carnegie Learning (2025). https://www.carnegielearning.com/ Cole, M. (1998). Cultural psychology: A once and future discipline. Harvard University Press. Dewey, John (1916). Democracy and Education: An Introduction to the Philosophy of Education. New York: Macmillan Farrell, H., Gopnik, A., Shalizi, C., & Evans, J. (2025a). Large AI models are cultural an...

arXiv 2025

[8] [2024]

and (2) ways to implement these technologies in our learning environments (Mollick, 2025a, 2025b) here I seek to promote a third way - a foundational approach to the above – principles to help us think about the productive uses of new technologies in education. These principles build on decades of scholarship of teaching and learning and of technology use...

1948