EMERGING RULE: Working Paper Series
The Governance Gap in K-12 Education:
Why the Most Consequential AI Challenge Is Pedagogical, Not Technical
June 2026
"The capacity of AI systems to be governed, audited, challenged, and held accountable depends on the existence of humans capable of performing those functions."
Suggested citation: Emerging Rule. (2026). The governance gap in K-12 education: Why the most consequential AI challenge is pedagogical, not technical.Emerging Rule Working Paper.
This working paper is circulated for discussion and comment. It has not undergone peer review. Views expressed are those of Emerging Rule and do not represent the positions of any institution cited herein.
Table of Contents
Executive Summary
I. Introduction: An Underappreciated Problem
II. The Structural Misalignment of Industrial-Era Education
III. The Demand-Side Failure: What Labor Markets Now Require
IV. What the Evidence Shows on Personalized Learning
V. AI Deployed in Education: Documented Risks of Poor Design
VI. A Framework for Responsible AI-Integrated Education
VII. The LevelShip Approach: Architecture and Principles
VIII. Implications: Policy, Governance, Workforce, and Investment
IX. Conclusion
Frequently Asked Questions
Suggested Data Visualizations
Pull Quotes for Publication
References
Appendix A: LevelShip Platform Architecture
Executive Summary
This paper argues that the most consequential AI governance challenge of the coming decade is not located in laboratory safety protocols, compute thresholds, or model alignment mathematics. It is located in schools.
The capacity of any society to benefit from increasingly powerful AI systems depends, in the first instance, on whether that society produces people who can think clearly, evaluate evidence, make sound judgments under uncertainty, and exercise meaningful oversight of automated systems. These capabilities are not instinctive. They are products of deliberate education. The dominant model of K-12 schooling, organized around fixed cohorts, standardized schedules, and retrospective assessment, was designed to produce reliable workers for structured industrial environments. It was not designed to produce the adaptive, critically literate decision-makers that AI-era institutions require.
The misalignment between what contemporary education systems produce and what AI-era economies, democracies, and governance structures require is measurable, widening, and underrepresented in the AI policy conversation.
OECD PISA 2022 data, spanning 81 participating countries, document declining mathematics performance alongside persistent equity gaps tied to socioeconomic background (OECD, 2023a). The World Economic Forum's Future of Jobs Report 2025, drawing on surveys of more than 1,000 global employers, found that approximately 39 percent of core workforce skills will be transformed or made obsolete by 2030, and that the skills gap is the leading barrier to business transformation, cited by 63 percent of employers surveyed (WEF, 2025). RAND Corporation research on personalized learning across 62 U.S. schools found positive achievement effects, particularly for the lowest-performing students, while also documenting significant implementation challenges and an evidence base that the researchers themselves described as incomplete (Pane et al., 2015; Pane, 2018).
These findings converge on the same structural question: whether K-12 education systems can evolve faster than the environments their graduates will inhabit.
This paper makes five arguments. First, the educational infrastructure inherited from the industrial era is structurally unsuited to the cognitive demands of AI-era participation. Second, personalized, adaptive learning, when grounded in evidence-based pedagogy and privacy-respecting data architecture, offers a credible pathway toward correcting that misalignment. Third, poorly designed AI deployment in education can actively impede learning and widen equity gaps, making responsible design a governance priority. Fourth, integrating student, educator, and family data into outcome-linked adaptive systems represents both a significant technical opportunity and an ethical obligation that demands careful institutional stewardship. Fifth, the most important variable in AI safety and governance discourse is one that rarely appears in it: what happens in classrooms.
The paper then examines how Emerging Rule's LevelShip platform addresses these challenges through its adaptive learning architecture, educator intelligence tools, family engagement infrastructure, and privacy-first data design. Emerging Rule acknowledges that peer-reviewed outcome evidence for LevelShip at scale does not yet exist. The platform's current case rests on its design architecture, its alignment with the existing evidence base, and its suitability for the institutional research partnerships necessary to generate rigorous evidence.
I. Introduction: An Underappreciated Problem
In the last three years, the governance of artificial intelligence has become one of the most intensively studied problems in public policy. Governments have commissioned reports, convened expert panels, passed legislation, and issued guidance. The European Union's AI Act, adopted in 2024, establishes a risk-based regulatory framework spanning prohibited uses, high-risk applications, and transparency obligations. The OECD AI Policy Observatory, UNESCO, and the United Nations have each published substantive governance frameworks addressing AI ethics, safety, and human rights. Academic institutions, think tanks, and research centers have produced a substantial body of literature on AI alignment, model safety, and the economic consequences of automation.
What is conspicuously absent from most of this literature is a sustained treatment of education as an AI governance variable.
The dominant implicit assumption in AI governance discourse is that the primary risks of advanced AI systems are located in the systems themselves. If models can be made sufficiently safe, interpretable, and aligned with human values, the argument runs, their deployment will be broadly beneficial. This premise is not unreasonable. It is, however, incomplete.
The capacity of AI systems to be governed, audited, challenged, and held accountable depends on the existence of humans capable of performing those functions. It depends on citizens who can evaluate AI-generated information critically, on workers who can collaborate with AI tools without surrendering independent judgment, on judges and legislators who understand what they are regulating, and on institutional leaders who can make sound decisions in environments where the consequences of error are increasing. These capabilities are cultivated through education systems. As such, strengthening the scale and quality of education is essential for meeting the needs of AI-era institutions.
The most important AI safety investment any society can make may be education of its citizens. This is a structural argument about the relationship between human capability and technological governance.
The paper proceeds as follows. Section II examines the structural design of industrial-era education and the specific ways it is misaligned with AI-era requirements. Section III analyzes the labor market evidence for this misalignment. Section IV reviews the evidence on personalized learning. Section V examines the risks of poorly designed AI deployment in education. Section VI proposes a framework for responsible AI-integrated education. Section VII describes the LevelShip approach. Section VIII draws out implications for policy, governance, workforce development, and institutional investment. Section IX concludes.
II. The Structural Misalignment of Industrial-Era Education
2.1 Origins and Design Logic
The organizational structure of contemporary K-12 schooling was largely codified in the late nineteenth and early twentieth centuries. Its defining features, grade levels organized by age cohort, fixed class periods, standardized curricula, periodic summative assessment, and a single teacher managing a heterogeneous group, reflect the administrative and economic priorities of the industrial era in which they were designed.
This design was coherent given its context. Industrial economies required workers who could follow instructions, maintain schedules, absorb technical training, and perform defined tasks reliably. Education systems organized around predictability and standardization served those requirements. The logic was not negligence; it was adaptation to genuine conditions.
Those conditions have changed substantially. The tasks that industrial-era schooling prepared students to perform are precisely the tasks that automation now performs most efficiently. Routine cognitive work, rule-based processing, information retrieval, and pattern recognition in structured environments are no longer predominantly human activities. What remains distinctively human, and increasingly valuable, is the capacity to exercise judgment in ambiguous situations, to evaluate the reliability of competing information sources, to engage ethically with complex decisions, and to adapt to circumstances that have no established precedent.
2.2 What Current Systems Measure
The persistence of industrial-era educational structures is reinforced by assessment systems that measure the outputs for which those structures were designed. Standardized testing, which dominates accountability frameworks across most OECD countries, measures reliable recall and application of defined knowledge within constrained time frames. These are not useless capabilities, but they represent a narrow subset of the competencies that contemporary institutions require.
The OECD's Future of Education and Skills 2030 project, which produced the Learning Compass 2030, identifies three categories of transformative competencies students require for productive participation in twenty-first-century society: creating new value, reconciling tensions and dilemmas, and taking responsibility (OECD, 2019). These competencies are qualitatively different from what standardized tests measure. They develop through sustained practice in unstructured problem-solving, exposure to genuine ethical complexity, and the experience of learning from failure in environments where failure has manageable consequences.
Most K-12 curricula provide limited systematic opportunities for developing these competencies. Assessment pressure redirects instructional time toward testable content, teacher effort toward examination preparation, and student attention toward score-maximization strategies rather than genuine comprehension. These are rational adaptations to the incentive structures accountability systems create. They are also structurally incompatible with developing the adaptive, critically reasoning capacities that AI-era participation demands.
2.3 The Equity Dimension
The structural limitations of industrial-era schooling fall unevenly across student populations. PISA 2022 findings across 81 participating countries document a persistent relationship between socioeconomic background and academic performance. Students from disadvantaged backgrounds receive, on average, less qualified instruction, less individualized attention, less access to enrichment activities, and less school-home communication support (OECD, 2023a).
This equity gap is not merely a matter of distributive justice, though it is that. It is also a human capital problem with direct implications for AI governance. If the capacities required to participate meaningfully in AI-era economies and institutions are distributed along socioeconomic lines, the result is a society in which the ability to govern, audit, and hold AI systems accountable is concentrated in exactly those populations whose interests AI systems are already most likely to serve. This structural concentration compounds existing inequalities in ways that democratic institutions will find increasingly difficult to manage.
2.4 The Pace Problem
Industrial-era institutions were characterized by relatively slow change. Competencies acquired in school remained relevant over careers measured in decades. The WEF Future of Jobs Report 2025 found that approximately 39 percent of core workforce skills are expected to be transformed or become obsolete by 2030, a five-year horizon shorter than a single K-12 cycle (WEF, 2025). A system that takes twelve years to produce a graduate and evaluates its success using assessments designed a decade ago is architecturally incapable of keeping pace with an economy whose skill requirements shift every five years.
III. The Demand-Side Failure: What Labor Markets Now Require
3.1 The Skills Gap as Institutional Failure
The WEF Future of Jobs Report 2025, based on surveys of over 1,000 employers representing more than 14 million workers globally, identifies the skills gap as the primary barrier to business transformation in the current period. Sixty-three percent of employers surveyed cited the inability to find workers with the skills they need as their leading constraint (WEF, 2025). This is a finding about institutional failure: the systems responsible for developing human capital are not producing what economies require.
The report projects that 170 million new roles will be created by 2030 while 92 million are displaced, for a net gain of approximately 78 million positions. Roles being created cluster around capabilities that require judgment, creativity, and the ability to work alongside AI systems productively. Roles being displaced cluster around routine cognitive and manual tasks.
3.2 The Cognitive Skills Premium
The fastest-growing skill category in the WEF 2025 data is cognitive skills, specifically analytical thinking and critical assessment, with seven in ten companies considering these essential in 2025. This is followed by resilience, flexibility, and leadership as socio-emotional competencies (WEF, 2025).
Two of the three fastest-growing skill categories are not technical in the narrow sense. They cannot be acquired by learning a programming language or completing a certification course. They develop through sustained practice in complex reasoning, deliberate exposure to ambiguity, and cultivation of intellectual habits over time. These are the competencies that educational systems have historically found most difficult to teach at scale, partly because they resist standardized measurement, and partly because industrial-era education was not designed with their development as a primary objective.
3.3 AI Literacy as a Governance Requirement
The third skill cluster, AI and digital literacy, encompasses more than the ability to use AI tools. It includes the capacity to evaluate AI outputs critically, to understand conditions under which AI systems are reliable or unreliable, to recognize algorithmic bias, and to participate meaningfully in institutional decisions about where and how AI systems are deployed. UNESCO's 2021 Recommendation on the Ethics of Artificial Intelligence specifically calls for member states to develop AI literacy as a prerequisite for meaningful human oversight and governance (UNESCO, 2021).
This form of AI literacy is not currently part of most national K-12 curricula. It is not assessed by existing standardized instruments. It is absent from most teacher preparation programs. A generation of students is entering AI-shaped labor markets and civic environments without the conceptual tools to navigate them responsibly.
3.4 Implications for Democratic Stability
Research in political science has consistently documented the relationship between educational attainment, critical reasoning capacity, and the quality of democratic participation. Citizens who lack the skills to evaluate information sources critically are more susceptible to misinformation, more vulnerable to manipulation, and less capable of meaningful participation in complex governance decisions.
AI systems substantially amplify the scale and sophistication at which information environments can be shaped. Algorithmic amplification of emotionally resonant content and AI-assisted disinformation are present features of contemporary information environments, not theoretical future risks. Democratic institutions that depend on informed citizen judgment face a structural problem if the educational systems responsible for producing that judgment are not producing the relevant capacities.
IV. What the Evidence Shows on Personalized Learning
4.1 The RAND Research Program
The most sustained body of empirical research on personalized learning at the K-12 level has been produced by the RAND Corporation, supported by the Bill and Melinda Gates Foundation. Their 2015 study examined achievement in 62 public charter and district schools implementing personalized learning practices, using a matched comparison group drawn from schools serving similar populations (Pane et al., 2015).
The key finding was that students in personalized learning schools made greater mathematics and reading progress over two school years compared to matched peers, and that students who entered with lower baseline performance made the most substantial relative gains. The schools achieving the largest achievement effects shared three implementation features: student grouping driven by data and responsive to individual needs, provision of data directly to students and inclusion of students in discussions about their own learning, and physical and organizational environments designed to support individualized instruction.
4.2 Epistemic Limitations
The honest assessment of the personalized learning evidence base is that it is encouraging but incomplete. RAND's own subsequent analysis, published in 2018, acknowledged that the field lacks clearly specified models with research evidence, and that designers of personalized learning programs will find important unanswered questions about which practices or combinations of practices are effective (Pane, 2018). The long-term effects of personalized learning on higher-order competencies, critical reasoning, adaptive judgment, ethical decision-making, have not been measured in any systematic way.
This epistemic gap is not a reason to abandon personalized learning. It is a reason to invest in the longitudinal, privacy-compliant research infrastructure necessary to generate better evidence. Organizations that make dramatic impact claims for personalized learning without the evidence base to support them should be viewed with appropriate skepticism.
4.3 Family Engagement: A Consistently Underutilized Lever
Fifty years of research on the relationship between family involvement and student academic achievement converge on a robust positive finding. Kim (2022), in a second-order meta-analysis synthesizing 23 prior meta-analyses encompassing 1,177 primary studies, found a consistent positive association between parental involvement and student academic achievement, with random-effects mean effect sizes of 0.18 for observational studies and 0.16 for intervention studies. The strongest associations were found for parent expectations and academic socialization, specifically parents communicating educational aspirations to their children and maintaining engagement with school activities.
These effect sizes are meaningful in educational research terms. More significant for the argument of this paper is that family engagement represents a lever for student outcomes that most educational technology ignores entirely. Most EdTech products are designed for the classroom or for individual students. The family, despite its consistent presence in research as a powerful predictor of outcomes, is routinely treated as a passive recipient of periodic reports rather than as an active participant in an adaptive learning system.
PISA 2022 findings add a cross-national dimension. Education systems with positive trends in parental engagement between 2018 and 2022 showed greater stability or improvement in mathematics performance during the pandemic disruption period (OECD, 2023b). The relationship between family engagement and learning resilience is not incidental.
V. AI Deployed in Education: Documented Risks of Poor Design
5.1 Cognitive Offloading
A growing body of research in cognitive science documents the risk of cognitive offloading in AI-assisted learning: the tendency for learners to delegate the mental effort that would otherwise drive learning to external tools, with the result that the learning itself does not occur. The productive struggle that accompanies genuine learning, the effortful process of working through a problem without immediate resolution, is a feature of effective learning, not a defect to be engineered away.
This concern is directly relevant to AI tutoring systems that provide immediate, high-quality outputs in response to student queries. When AI assistance eliminates the struggle that drives learning, it may produce short-term performance improvements on practice tasks while undermining the development of independent competence. This effect is particularly consequential for students in early stages of building foundational knowledge, whose capacity to evaluate AI outputs critically depends on having acquired the foundational knowledge in the first place.
5.2 Equity and the Access Problem
AI tools in education are not neutral with respect to equity. Students from higher-income families are more likely to have reliable broadband access, current devices, and adult supervision that supports productive AI tool use. Students from lower-income families are more likely to encounter AI tools in under-resourced school settings, without the pedagogical frameworks that support productive use. AI deployment in education that proceeds without deliberate equity design is likely to widen existing achievement gaps rather than close them.
This concern extends to the AI training data problem. Systems trained predominantly on the output of high-resource, English-speaking, economically privileged users will tend to reflect the assumptions, values, and knowledge patterns of those users. When these systems are deployed in diverse educational contexts without adjustment, they risk systematically undeserving students whose backgrounds are not represented in the training corpus.
5.3 Privacy Governance in K-12 Contexts
K-12 students are among the most legally protected data subjects in the United States and in a growing number of jurisdictions internationally. The Family Educational Rights and Privacy Act (FERPA), the Children's Online Privacy Protection Act (COPPA), and state-level statutes establish significant restrictions on the collection, use, and sharing of student data.
AI learning systems that collect student behavioral data, learning pattern data, and family engagement data must treat compliance with these frameworks as a design requirement from the outset. Systems built without this architecture face not only legal exposure but also a more fundamental erosion of institutional trust. Families who believe their children's data is being collected without appropriate consent or protection will, rationally, resist the technologies that require that data. Building the trust that effective AI-integrated education requires depends on demonstrating that privacy protection is an architectural priority.
The UNESCO 2021 Recommendation on the Ethics of Artificial Intelligence identifies the right to privacy and data protection as among its ten foundational principles, with particular attention to contexts involving children (UNESCO, 2021). The OECD AI Policy Observatory similarly identifies privacy and data governance as central requirements for trustworthy AI in high-risk applications, a category that includes AI in education.
VI. A Framework for Responsible AI-Integrated Education
6.1 Five Guiding Principles
Drawing on the UNESCO Recommendation on the Ethics of Artificial Intelligence (2021), the OECD Learning Compass 2030 (OECD, 2019), and the RAND Corporation's personalized learning research (Pane et al., 2015; Pane, 2018), this section proposes five principles for responsible AI-integrated K-12 education.
First: Human agency as the non-negotiable anchor.
AI systems in education must augment human thinking rather than displace it. This means designing for productive cognitive engagement rather than cognitive offloading, for developing independent judgment rather than AI reliance, and for maintaining the productive struggle that genuine learning requires. The measure of a sound AI learning system is not whether it makes learning feel easier in the short term but whether it develops the student's independent capacity over time.
Second: Privacy by design, not by compliance.
Student data collection in AI-integrated education must meet a standard higher than regulatory compliance. It must meet a standard of demonstrable benefit to the student whose data is collected. Data architecture must embed consent, transparency, and meaningful parental oversight from the design stage.
Third: Equity as a design constraint.
AI-integrated education systems must be evaluated against equity outcomes from the outset. If a system improves aggregate outcomes while widening the gap between advantaged and disadvantaged students, it has failed in a fundamental respect. Equity must be built into algorithmic design, training data selection, access infrastructure, and outcome measurement.
Fourth: Whole-system thinking.
A student's educational development occurs across multiple environments: the classroom, the home, peer relationships, and the community. Educational technology that addresses only the classroom component addresses only a fraction of the system that produces learning outcomes. Responsible AI-integrated education must be designed with the full system in mind, connecting teacher intelligence, family engagement, and student agency within a coherent framework.
Fifth: Evidence before scale.
The history of educational technology includes instances of large-scale deployment of systems whose effectiveness was assumed rather than demonstrated. Responsible deployment requires rigorous evidence generation alongside implementation, not in retrospect. RAND's research is explicit that the personalized learning evidence base is still developing, and that scaling should be accompanied by continued investment in rigorous evaluation (Pane, 2018).
6.2 The Research Infrastructure Gap
Applying these principles reveals a structural problem in current EdTech development. The research infrastructure needed to generate rigorous, longitudinal, privacy-compliant evidence of learning outcomes in AI-integrated K-12 settings does not currently exist at meaningful scale. IRB-compliant research partnerships between EdTech developers, school districts, and universities are possible but uncommon. The data sharing agreements, consent frameworks, and de-identification protocols that would allow credible outcome studies are not standardized across the field.
Closing this gap requires institutional investment from research organizations and government alongside a commitment from EdTech developers to treat rigorous evidence generation as a precondition for scale, not an afterthought.
VII. The LevelShip Approach: Architecture and Principles
7.1 Organizational Context
Emerging Rule is a K–12 educational technology public benefit corporation founded in 2016 and recognized by InnoStars, 1776, StartEd at NYU, Singularity University, the Silicon Valley Innovation and Entrepreneurship Forum (SVIEF), and the U.S.–China Innovation Program. Its flagship platform, LevelShip, is designed around the five principles described in Section VI. The following description draws on publicly available materials consistent with the disclosure of Emerging Rule’s intellectual property. Emerging Rule acknowledges that peer-reviewed evidence of outcomes for LevelShip at scale is not yet available. At present, evaluation of the platform is grounded in its design architecture, its pedagogical rationale, and its alignment with research on personalized learning and family engagement reviewed in Sections IV and V.
7.2 Adaptive Learning Architecture
LevelShip's core function is the continuous identification, classification, and developmental allocation of each student's emerging competencies toward an optimized career trajectory. The system processes signals from student responses, time-on-task patterns, error types, and progression rates, using these inputs not only to adjust content difficulty and sequencing in real time, but to build a longitudinal competency profile that maps each student's demonstrated strengths to viable and well-matched vocational and academic pathways.
This approach draws on established theory in career development psychology, particularly Holland's (1997) person-environment fit model and Super's (1980) life-span developmental framework, both of which hold that meaningful career outcomes depend on early, accurate identification of individual aptitudes and their alignment with compatible environments. LevelShip operationalizes this principle at the K-12 level by treating each instructional interaction as a data point in an ongoing competency mapping process rather than a discrete performance event to be graded and archived.
The platform is designed to distinguish between surface-level performance gaps, where a student answers questions incorrectly within a unit, and structural comprehension gaps, where a student lacks the conceptual prerequisite for the skill being assessed. These are pedagogically distinct conditions. The former may indicate a need for reinforcement or additional practice within a competency domain already within the student's developmental reach. The latter indicates that the student's competency profile has not yet developed the foundational architecture on which the target skill depends, and that premature progression will compound rather than resolve the gap. Conflating these two conditions, as periodic summative assessment routinely does, produces instructional responses that address surface symptoms while leaving structural gaps intact.
The distinction also carries direct implications for career pathway allocation. A student whose competency profile shows persistent surface-level gaps in quantitative reasoning may be misclassified as unsuited to STEM pathways when the underlying issue is an unresolved foundational gap in number sense or proportional reasoning that targeted intervention could close. Early and accurate differentiation between these gap types is therefore consequential not only for immediate instructional design but for the longer-term career development trajectories that early educational experiences set in motion. Whether the platform's AI architecture reliably makes this distinction in practice is a question that independent longitudinal evaluation will need to answer. In this respect, LevelShip does not narrow a student's vocational horizon by early classification; it expands it, ensuring that as AI continues to reshape the skill requirements of existing occupations and generate new ones, each student's competency profile is continuously refined to reflect their demonstrated strengths, developmental progress, and the evolving demands of the economic and civic environments they will inhabit.
7.3 Educator Intelligence Layer
LevelShip provides educators with a real-time dashboard displaying each student's performance trajectory. The design intention is to shift teacher attention from retrospective grading toward prospective intervention: identifying and addressing learning difficulties before they compound rather than reacting to failures after they have occurred.
This addresses a well-documented challenge in classroom instruction. In a class of twenty-five to thirty students operating without data support, a teacher will typically identify students who are significantly below grade level, and may identify students who are exceeding expectations, but will have limited visibility into the trajectories of students in the middle of the distribution. Trajectory data, as distinct from snapshot performance data, provides a qualitatively different informational basis for instructional decisions.
7.4 Family Engagement Infrastructure
LevelShip includes a family-facing interface providing parents and guardians with real-time progress updates translated into plain language, alongside specific guidance on how families can support learning at home. The rationale is grounded in the meta-analytic evidence reviewed in Section IV: the positive relationship between family engagement and student outcomes is one of the most consistently replicated findings in K-12 education research (Castro et al., 2015; Kim, 2022). A system that extends intelligence and agency to families addresses a gap that the vast majority of existing EdTech products do not.
The privacy governance requirements of this design feature deserve explicit acknowledgment. Engaging families in a student's learning data requires careful consent design, clear data use policies, and robust security practices. The benefit of family engagement must be achieved without compromising the student's educational privacy or the family's right to informed consent.
7.5 The Outcome-Linked Data Architecture
LevelShip's most distinctive architectural feature is its design intention to link student, teacher, and family engagement data to learning outcomes. If implemented with appropriate privacy protections and research governance, this would create an unusual dataset: one capturing not just what students did, but which patterns of student behavior, educator response, and family engagement are predictive of specific learning outcomes.
Most EdTech platforms collect engagement data. Few attempt to link these systematically to outcomes in ways that would allow the patterns to be analyzed for predictive value. A platform that generates this linkage at scale, under appropriate de-identification and research governance, would represent a meaningful contribution to the evidence base for personalized learning. Realizing this potential requires institutional partnership with university research centers capable of leading IRB-compliant outcome studies and data sharing frameworks that meet FERPA's research exception requirements.
7.6 What LevelShip Claims and Does Not Claim
LevelShip claims to have designed an adaptive learning platform whose architecture is consistent with the evidence base for personalized learning, grounded in sound pedagogical reasoning, and built with a privacy-first approach to student data. It claims to have identified a gap in the EdTech landscape, specifically the absence of whole-system approaches connecting student, educator, and family intelligence within a unified platform, and to have built an architecture designed to address that gap.
LevelShip does not claim peer-reviewed evidence of learning outcomes at scale. It does not claim its AI architecture has been independently validated. It invites the institutional scrutiny and research partnership necessary to generate that evidence.
VIII. Implications: Policy, Governance, Workforce, and Investment
8.1 Policy Implications
Accountability frameworks should measure the competencies that AI-era participation requires, not only those that industrial-era assessment can efficiently capture.
Education policymakers at national and subnational levels should consider the following priorities. Accountability frameworks should measure the competencies that AI-era participation requires. The OECD Learning Compass 2030 provides a coherent framework of transformative competencies, including creating new value, reconciling tensions and dilemmas, and taking responsibility, that could inform assessment reform (OECD, 2019).
Student privacy regulation, while necessary, should be designed in ways that do not preclude the generation of evidence that responsible AI-integrated education requires. FERPA's research exception, which permits de-identified educational data to be used in formally structured research, provides a workable framework. Policymakers should invest in standardizing the data sharing agreements and IRB frameworks that would allow this exception to be used more systematically.
Teacher preparation programs should incorporate AI literacy, formative assessment practice, and data-driven instructional tools as core components. UNESCO's 2021 guidance for policymakers on AI in education specifically recommends developing teacher competency in AI tools as a prerequisite for effective implementation (UNESCO, 2021).
Public investment in K-12 educational research infrastructure should be understood as AI governance investment. The national interest in producing citizens capable of navigating AI environments is directly served by improving the rigor and quality of K-12 education research.
8.2 AI Governance Implications
Organizations engaged in AI governance, including the OECD AI Policy Observatory, UNESCO's AI Ethics Division, and national AI advisory bodies, should incorporate education explicitly as a governance variable. Current frameworks focus primarily on the design, deployment, and accountability of AI systems. They give insufficient attention to the human capability requirements for effective AI governance to function.
Specifically, AI governance frameworks should address AI literacy standards in K-12 curricula, the governance requirements for AI systems deployed in educational settings involving minors, and the relationship between educational quality and the broader capacity for democratic oversight of AI systems. The European Union's AI Act classifies AI systems used in education as high-risk applications requiring conformity assessments, transparency obligations, and human oversight mechanisms. Its implementation should be informed by the pedagogical and privacy considerations this paper has outlined.
8.3 Workforce and Economic Development Implications
The skills gap identified by the World Economic Forum is not a problem that can be resolved through employer-sponsored reskilling programs alone. The cognitive and socio-emotional competencies that employers most urgently need, analytical thinking, resilience, adaptive reasoning, ethical judgment, develop over years, beginning in early education. Economic development strategies that prioritize AI adoption without corresponding investment in educational quality are likely to find that the human capital required to realize AI's productive potential is absent.
This dynamic reinforces the case for treating K-12 educational quality as economic infrastructure rather than social expenditure. The return on investment in education, extensively documented in economics literature, is likely to increase as AI transforms labor markets and raises the premium on capabilities that resist automation.
8.4 Investor Implications
The EdTech investment market is currently characterized by information asymmetry. Platforms making confident impact claims attract capital, while platforms that acknowledge the genuine complexity of measuring educational outcomes are at a short-term disadvantage. This dynamic creates selection pressure toward overconfident claims and rapid deployment, precisely the conditions that the RAND research suggests are associated with implementation failure.
Investors with genuine interest in long-term value creation in the EdTech sector should apply a different evaluative standard: the quality of a platform's pedagogical reasoning, the rigor of its privacy architecture, the coherence of its approach to evidence generation, and the strength of institutional partnerships that would enable rigorous outcome evaluation. Platforms meeting these criteria are more likely to be durable, trusted by institutional partners, and ultimately impactful.
The international dimension of this opportunity deserves acknowledgment. Emerging Rule's presence in Latin American markets, through its Spanish-language interface and regional leadership, provides a pathway for deployment in contexts where privacy frameworks may permit more transparent outcome documentation in the near term. Well-documented international implementations can generate early evidence and demonstrate platform adaptability across diverse educational contexts.
8.5 Education Sector Implications
School districts and education administrators should consider applying the five principles outlined in Section VI when evaluating any AI-integrated learning platform: human agency protection, privacy by design, equity as a design constraint, whole-system thinking, and evidence before scale. Districts that adopt platforms failing to meet these criteria are accepting institutional and reputational risks that may be difficult to manage after adoption has occurred.
Districts positioned to participate in IRB-compliant outcome research, either through existing university partnerships or through developing new ones, should treat this as a strategic opportunity. The districts that generate credible evidence of effective AI-integrated learning will influence the policy and investment decisions that shape this field.
IX. Conclusion
Artificial intelligence does not organize itself around the assumptions that have historically governed human education. Its foundational architecture, statistical inference across large datasets, optimization through iterative mathematical processes, and pattern recognition operating at a scale and speed no human institution can replicate, is indifferent to the curricular traditions, credentialing hierarchies, and pedagogical orthodoxies that educational systems have accumulated over centuries. Where human education has largely operated on normative consensus, philosophical tradition, and political compromise, AI operates on empirical signals. It does not grade on a curve; it identifies patterns in data and optimizes toward defined objectives with a consistency and scalability that expose, rather than accommodate, the structural inefficiencies of systems built around the average student, the fixed academic year, and the standardized examination.
Rather than framing this as an argument that AI is superior to human judgment, this work focuses on AI’s functional logic. That logic is grounded in continuous measurement, adaptive optimization, and outcome-linked feedback, and it represents a structural shift for educational systems whose operating assumptions have remained largely unchanged since the industrial era. What remains an open institutional question is whether educational systems will adapt their underlying architecture in time. The evidence reviewed in this paper suggests the gap between what those systems currently produce and what AI-integrated societies require is not approaching; it is already present and widening. At what point a policy challenge becomes a governance crisis is a matter of threshold, not trajectory.
The governance of artificial intelligence is a problem of institutional design. Institutions govern AI systems through the judgment of the people who staff them, the frameworks those people understand, and the values those people have internalized. If the people staffing institutions are not equipped to evaluate AI systems critically, to recognize their failure modes, to exercise meaningful oversight, and to make sound decisions about their deployment, then governance frameworks, regardless of their technical sophistication, will be insufficient.
Educational systems were designed, with considerable success for their context, to produce workers and citizens for industrial-era economies and institutions. That context has changed faster than educational architecture typically adapts, and the gap between what education systems currently produce and what AI-era institutions require is measurable and widening.
Existing educational institutions carry enormous accumulated knowledge, professional commitment, and social infrastructure. The task is to redesign their operating logic: to shift from fixed cohort instruction to individualized progression, from summative to continuous assessment, from classroom-only intervention to whole-system engagement, and from standardized content delivery to the deliberate cultivation of adaptive reasoning.
This redesign requires pedagogical change, institutional commitment, research investment, and time. Technology that provides teachers with intelligence to intervene early, families with visibility to support learning at home, and students with pathways calibrated to their individual development can accelerate that process. Technology deployed absent these conditions can impede it, by eliminating the productive struggle that learning requires, concentrating benefits among already-advantaged populations, or eroding institutional trust through inadequate privacy governance.
LevelShip, developed by Emerging Rule, is an attempt to build technology that supports this redesign. Whether it succeeds is a question that rigorous evidence will answer. The contribution this paper seeks to make is to articulate the framework against which any answer should be assessed.
The most important AI safety investment is not in the laboratory. It is in the classroom.
The cost of delay is not theoretical. The OECD's 2022 assessment of 81 countries, conducted before generative AI had reached its current scale of deployment and social integration, documents declining mathematics performance and widening socioeconomic achievement gaps at precisely the moment AI systems are raising the cognitive threshold for meaningful economic and civic participation. The World Economic Forum projects that 39 percent of core workforce skills will be transformed or obsolete by 2030, a deadline that falls within the current K-12 cycle of students already enrolled in classrooms today. The meta-analytic evidence on family engagement, spanning 50 years and 1,177 primary studies, demonstrates that the infrastructure for improving outcomes exists and remains chronically underutilized.
Institutions that defer the redesign of their educational architecture until the consequences are undeniable will find that the populations hardest hit by that deferral are the ones least equipped to recover from it. The students sitting in classrooms this year will govern, work within, and be governed by AI systems whose sophistication will exceed anything currently deployed. Whether they are prepared to do so with judgment, critical capacity, and ethical literacy is a decision that educational institutions, policymakers, and investors are making right now, through action or through inaction.
Frequently Asked Questions
What do you do?
LevelShip identifies what each student is genuinely good at, tracks how their abilities develop over time, and connects that profile to the career paths most likely to suit them, while keeping teachers and families informed at every step so no one is working in the dark.
Why does an EdTech company's working paper address AI governance and democratic stability?
Because the argument is that these domains are structurally connected. The capacity of institutions to govern AI depends on the capability of the people staffing them. That capability is produced by education. An organization building educational technology for K-12 students has a direct stake in articulating this connection clearly, and an obligation to subject that articulation to scrutiny.
Does LevelShip have peer-reviewed evidence of learning outcomes at scale?
No. This paper is explicit on this point. The platform's current evidence base consists of its design architecture, its pedagogical reasoning, and its alignment with established research on personalized learning and family engagement. Peer-reviewed outcome evidence requires institutional research partnerships that are in development. Investors and institutional partners should evaluate the platform accordingly.
How does LevelShip protect student privacy?
LevelShip is designed with FERPA and COPPA compliance as a baseline, with a stated design intention of building privacy protections into the data architecture from the outset. Technical implementation details and compliance documentation are available from Emerging Rule directly. Independent third-party privacy audit would be an appropriate condition for any institutional adoption.
What distinguishes LevelShip from other adaptive learning platforms?
The primary architectural distinction is whole-system design, connecting student learning data, teacher instructional intelligence, and family engagement within a unified platform. Most adaptive learning systems address one or at most two of these components. The outcome-linked data architecture, designed to capture which patterns of behavior across all three participant types are associated with specific learning outcomes, is a second distinguishing feature. Whether these distinctions translate into superior learning outcomes is a question for rigorous evaluation.
What would constitute credible evidence of LevelShip's effectiveness?
Rigorous evidence would require a randomized or quasi-experimental design with a matched comparison group, longitudinal outcome measurement over at least two academic years, independent measurement by researchers not affiliated with Emerging Rule, IRB approval and formal data sharing agreements with participating districts, and publication in a peer-reviewed venue. This is the standard applied to any educational intervention claiming significant effects.
Appendix A: Supporting Figures
Pull Quotes for Publication
The capacity of AI systems to be governed, audited, challenged, and held accountable depends on the existence of humans capable of performing those functions.
A system that takes twelve years to produce a graduate and evaluates its success using assessments designed a decade ago is architecturally incapable of keeping pace with an economy whose skill requirements shift every five years.
Two of the three fastest-growing skill categories in the WEF 2025 data are not technical in the narrow sense. They develop through sustained practice in complex reasoning, deliberate exposure to ambiguity, and cultivation of intellectual habits over time.
If AI deployment in education proceeds without deliberate equity design, it is likely to widen existing achievement gaps rather than close them.
The most important AI safety investment is not in the laboratory. It is in the classroom.
References
All references verified as of June 2026.
Castro, M., Expósito-Casas, E., López-Martín, E., Lizasoain, L., Navarro-Asencio, E., & Gaviria, J. L. (2015). Parental involvement on student academic achievement: A meta-analysis. Educational Research Review, 14, 33–46. https://doi.org/10.1016/j.edurev.2015.01.002
Kim, S. W. (2022). Fifty years of parental involvement and achievement research: A second-order meta-analysis. Educational Research Review, 37, 100463. https://doi.org/10.1016/j.edurev.2022.100463
OECD. (2019). OECD Learning Compass 2030: A series of concept notes. Organisation for Economic Co-operation and Development. https://www.oecd.org/education/2030-project/
OECD. (2023a). PISA 2022 results (Volume I): The state of learning and equity in education. OECD Publishing. https://doi.org/10.1787/53f23881-en
OECD. (2023b). PISA 2022 results (Volume II): Learning during – and from – disruption. OECD
Holland, J. L. (1997). Making vocational choices: A theory of vocational personalities and work environments (3rd ed.). Psychological Assessment Resources.
Publishing. https://doi.org/10.1787/a97db61c-en
Pane, J. F., Steiner, E. D., Baird, M. D., & Hamilton, L. S. (2015). Continued progress: Promising evidence on personalized learning. RAND Corporation. https://doi.org/10.7249/RR1365
Pane, J. F. (2018). Strategies for implementing personalized learning while evidence and resources are underdeveloped. RAND Corporation. https://doi.org/10.7249/PE314
Super, D. E. (1980). A life-span, life-space approach to career development. Journal of Vocational Behavior, 16(3), 282–298. https://doi.org/10.1016/0001-8791(80)90056-1
UNESCO. (2021). Recommendation on the ethics of artificial intelligence. United Nations Educational, Scientific and Cultural Organization. https://www.unesco.org/en/articles/ai-and-education-guidance-policy-makers
World Economic Forum. (2025, January 8). Future of jobs report 2025. World Economic Forum. https://www.weforum.org/publications/the-future-of-jobs-report-2025/
Appendix A: LevelShip Platform Architecture
The following description is based exclusively on publicly available platform materials and information. It does not include, disclose, or rely upon our intellectual property, proprietary information, confidential information, trade secrets, or other non-public materials. It is provided solely as a technical reference for readers evaluating the platform's stated design against the principles discussed in Section VI and should not be interpreted as an independent verification of the platform's capabilities, performance, security, privacy practices, or legal compliance.
A.1 Adaptive Engine
LevelShip's adaptive engine processes continuous behavioral signals from student interactions: response accuracy, response latency, error pattern analysis, and progression through content sequences. The engine makes pathway adjustments in real time as students move through learning tasks, rather than waiting for scheduled assessment events. The eight-step algorithmic process involves data collection, preprocessing, feature engineering, model selection, model training, model evaluation, skill allocation, and continuous iteration, with the stated goal of allocating specific skills to individual learners based on performance metrics and available demographic context.
A.2 Educator Dashboard
The educator-facing interface presents each student's learning trajectory in a real-time dashboard format. The stated design intention is to make each student's current position and direction of movement visible to the teacher, providing an informational basis for timely intervention before difficulties compound. The dashboard is designed to provide actionable intelligence rather than raw data requiring manual interpretation.
A.3 Family Interface
The family-facing interface translates platform data into plain-language progress summaries and actionable guidance for home-based learning support. The design intention is to make family members active participants in the learning system rather than passive recipients of periodic report cards. A Spanish-language interface, Protagonistas, is available, reflecting Emerging Rule's engagement with Latin American school systems.
A.4 Privacy Architecture
LevelShip is designed to align with FERPA and COPPA requirements as baseline privacy and data protection standards. Its data architecture separates individually identifiable student information from aggregated or de-identified data that may be used for platform improvement and outcome analysis. The extent to which these practices satisfy applicable legal requirements, including any FERPA exceptions, has not been independently verified by publicly available technical audits. Institutions considering adoption should request and review detailed technical, legal, and compliance documentation and conduct their own due diligence regarding privacy, security, and regulatory compliance.
For further evaluation or inquiries, contact us.
Contact Page | emergingrule.com | admin@emergingrule.com
This working paper is available for reproduction with attribution. Peer review correspondence is welcomed.