Introduction: The reshaping of the educational technology paradigm and the critical period of human development
Over the past decade, the exponential evolution of artificial intelligence (AI) technology has fundamentally reshaped the global education ecosystem. From early adaptive learning platforms to today's generative artificial intelligence (GenAI) with multi-modal interaction, natural language processing and emotional computing capabilities, technical systems have been deeply embedded in children's family life, school education and social interactions. Today in 2026, what we face is no longer the introduction of a single tool, but a new digital living environment driven by algorithms and deeply involved in human cognition and emotion. 4.
Research in developmental psychology and neurobiology has conclusively pointed out that the period from birth to approximately 25 years old is the most critical and sensitive period for human brain development and neuroplasticity 6 . During this period, the nature of external environmental stimuli will directly shape children's cognitive structures, socioemotional skills, and behavioral patterns 6 . Currently, the discussion on the role of AI in education has moved beyond simple instrumental theory to a comprehensive examination of its role in cognitive science, social psychology, ethics, and law.
On the one hand, AI has proven to be highly effective in providing high-frequency personalized tutoring, special education intervention, and breaking down barriers to traditional educational resources 8 . On the other hand, the deep risks it causes in terms of cognitive offloading, anthropomorphic attachment, algorithmic bias, and data privacy force policymakers and educators to re-evaluate the boundaries of "human-machine collaboration" 10 . This report is based on a comprehensive examination of the latest empirical research, multi-national policy frameworks and technological evolution paths. It aims to systematically analyse the substantive impact of artificial intelligence on children in the field of education and reveal the causal relationship and future evolution trends behind the data representation.
The deep reconstruction and double-edged sword effect of artificial intelligence on children’s cognitive development
The in-depth application of artificial intelligence in educational environments has triggered profound discussions about children’s cognitive mechanisms. Unlike traditional search engines or educational software, generative AI can directly output highly condensed and complex results. This feature is changing the neuropsychological process of children's acquisition, processing and internalization of knowledge.
The “Learning-Performance Paradox” and the Risk of Cognitive Offloading
In the 247-page "Digital Education Outlook" report released by the Organization for Economic Cooperation and Development (OECD) in 2026, it revealed a core phenomenon that alarms the education community: the "Learning-Performance Paradox" 13. Large-scale empirical studies and field experiments show that when students use general generative AI tools, their short-term task scores (such as essay writing, problem solving) can increase by up to 48%; however, when AI assistance is removed in subsequent tests, the actual performance of these students drops by 17% 11.
This stark contrast in data reveals a key mechanism in cognitive science: the abuse of cognitive offloading 8 . Deep, lasting learning relies on retrieval practice (Retrieval Practice), cognitive refinement, and problem-solving attempts under uncertainty11. When the AI system directly provides completed output (such as drafting text, comprehensive answers), the cognitive friction (Cognitive Friction) that should be completed internally by the learner is replaced by external algorithms11. For novice learners who lack basic knowledge structure, this "outsourcing" will lead to serious "metacognitive laziness" (Metacognitive Laziness) 15. Research data shows that under independent writing conditions, students’ recall rate of learned knowledge can reach 89%, but when relying on general generative tools such as ChatGPT, the recall rate plummets to only 12% 13 . This shows that over-reliance on AI tools not only fails to promote deep learning, but also weakens the neural basis for children to build long-term memory and critical thinking 4.
From a neurobiological perspective, overcoming intellectual challenges activates the brain’s reward system, producing dopamine and reinforcing intrinsic motivation to learn (i.e., the “effort-reward” cycle) 16 . The instant answers generated by AI in a few seconds bypass this natural entrance to learning, depriving children of the opportunity to gain this inner sense of accomplishment during exploration, causing the learning experience to degrade from active knowledge construction to a passive consumption process, which may damage children's intrinsic love for the learning process itself in the long run 16 .
Epistemological overtrust, automation bias, and the superiority of human cognition
In addition to blunting cognitive abilities, AI also profoundly affects children’s epistemological tendencies (Epistemic Tendencies). Due to limitations in their development stages, children have not yet fully formed a complete critical evaluation ability, and are easily prone to "automation bias" and "overtrust" in machine output. 6 Dual-Process Theory points out that trust judgments originate from the interaction between the fast and intuitive "System 1" and the slower "System 2" responsible for evaluating reliability and transparency. 18
In the interaction between children and AI, because AI often displays a high degree of fluency and authority, children can easily skip the analysis process of "System 2" and develop epistemological blind obedience. Research points out that as interaction experience increases, "overtrust" may evolve into a lasting epistemological habit, leading to children's lack of "epistemic vigilance" (Epistemic Vigilance) when facing complex information 19 . Furthermore, overreliance on “computational thinking” may cause education systems to ignore the unique strengths of human cognition. Neuroscientific research by scholars such as Tina Grotzer of the Harvard School of Education has strongly demonstrated that although human thinking contains Bayesian calculation processes, it is "Better than Bayesian" in many aspects20. For example, humans’ somatic markers enable us to make rapid intuitive leaps; in game experiments, kindergarten children are able to use strategic information to make faster and more informed decisions than purely Bayesian computational methods. 20 If the education system blindly emphasizes linear computing logic led by AI, it may inhibit the development of children's embodied cognition and intuitive creativity 20.
Socio-emotional development, anthropomorphic mechanism and ecological reconstruction of human-computer interaction
With the popularization of educational robots and the application of affective computing (Affective Computing) technology, children's social interaction objects are expanding from a single companion/adult to a new digital ecosystem including artificial intelligence. While this shift promotes emotional engagement in children, it also raises systemic concerns about anthropomorphic attachment, loss of authentic social skills, and a surge in screen time.
The Intertemporal Evolution of Anthropomorphism and the Projection of Theory of Mind (ToM)
Modern educational robots (such as Miko, Moxie, ROYBI, etc.) are not only designed as computing tools to transfer knowledge, but also are shaped into "social partners" with personality and capable of facial recognition and emotional training 21 . Research shows that children’s degree of anthropomorphism (Anthropomorphization) of robots is closely related to their age and cognitive development stage25. In the preschool stage (such as 3 to 5 years old), children are more inclined to attribute biological properties to robots and subconsciously apply their developing "Theory of Mind" (ToM), that is, they believe that robots have beliefs, desires and hidden emotions 25 .
In a Property Projection Task, 3-year-olds were more likely than 5-year-olds to think that a humanoid robot was "alive" and to attribute biological characteristics to it27. However, this kind of anthropomorphic trust based on surface cues has complex evolutionary dynamics. Empirical research on children's trust strategies shows that children evaluate information providers based on their past accuracy when deciding who to trust 29 . Generational differences are particularly evident here: compared to older children, younger children show higher levels of trust even when faced with unreliable humans than with unreliable robots 29 . This shows that children’s social trust mechanism still has a strong species preference in the early stages of development, and their trust in robots (divided into social trust and capability trust) is unstable 30 .
Long-term companion AI applications do demonstrate emotional value in specific areas. Long-term follow-up research based on Domestication Theory shows that PARO robots used to treat neurodevelopmental disorders (such as autism) can effectively reduce children's anxiety, increase positive emotions, and promote a deep human-robot emotional bond 31. But in the general educational context, this bond may evolve into a one-way "parasocial relationship". Since robots lack the subtle details and physiological synchronization of real human emotions (biological synchronization, such as heart rhythm, gaze patterns, and body temperature), children who have been immersed in human-computer interaction for a long time may lose the ability to interpret complex human social cues. 6 In addition, the emergence of “Robot Abuse” has raised new ethical concerns. For example, the Shelly robot turtle, designed to teach children about gentle interactions (it retreats into its shell when hit with too much force), revealed that children may develop negative behaviors rather than virtues when interacting with painless AI entities, hindering their moral and empathic development 32 .
Surge in screen time and alienation from real social relationships
AI-powered apps and devices have significantly increased children’s exposure to digital devices. According to the latest research from the Children's Hospital of Orange County (CHOC) in California, the average daily screen time among teenagers has reached a staggering 8 hours, and the time spent on devices by pre-teens (8 to 12 years old) has also reached 5.5 hours, accounting for half of their waking hours 33 .
| The negative cascading effects of screen time on children’s physical and mental health | Specific manifestations and mechanisms | Source basis |
|---|---|---|
| Physical health damage | The risk of myopia increases sharply, affecting the normal development of the eyeballs; increased sedentary behaviour leads to metabolic problems. | 6 |
| Neural and sleep rhythm interference | Screen blue light and high-frequency interactive algorithms destroy melatonin secretion, causing sleep disorders and circadian rhythm disorders. | 4 |
| Degeneration of Attention and Executive Function | Attention-capturing algorithms overstimulate the dopamine system and destroy the ability to delay gratification. | 6 |
| Crowding out of real interpersonal relationships | Reduce the time of physical interaction with peers, parents and real society, and increase depressive symptoms and loneliness. | 4 |
The anthropomorphic characteristics of AI chatbots may have a strong crowding-out effect on adolescents’ real interpersonal relationships. A survey by the Centre for Democracy and Technology (CDT) for the 2024-2025 school year showed that 42% of students used AI to seek mental health support, 42% used it as a friend or partner, 19% even tried to have a romantic relationship with it, and 42% tried to escape real life through AI 10 . This high-frequency “human-computer bond” cuts off the way for children to build healthy personalities through peer friction, conflict resolution and deep empathy 6 . Ying Xu, an assistant professor at Harvard University, emphasized that although AI can ask questions while reading to improve vocabulary, it absolutely cannot replicate the deep relationship building and sociolinguistic development that human interactions bring 35.
Breakthrough in personalized intervention for special education and “long-tail learners”
Of all education segments, Special Education and Neurodiversity support are among the areas with the most transformative potential for AI. A report from the U.S. Department of Education pointed out that traditional standardized courses can usually only serve "Teaching to the middle", while the emergence of AI provides the possibility to solve the "Long Tail of Learner Variability" 3. By breaking the “one-size-fits-all” teaching model, AI provides unprecedented educational accessibility for children with learning disabilities (LD), attention deficit hyperactivity disorder (ADHD), autism spectrum disorder (ASD), and visual and auditory impairments 8 .
Precision coaching tools for neurodiverse children
For children with dyslexia (Dyslexia), AI-based tools are reshaping the way information is processed. For example, Mindgrasp.ai, a mainstream tool in 2025, can automatically convert dense academic materials into summaries, drawn cards, and customized tests, thereby reducing cognitive load; VoiceType AI provides a voice dictation function with an accuracy of up to 99.7%, overcoming writing difficulties; Helperbird allows users to customize the fonts, colors, and overlays of the reading environment39. At the same time, Microsoft's Immersive Reader, equipped with modern text-to-speech (TTS) technology, can adjust the speaking speed according to the complexity of the content and even identify potentially difficult words to provide contextual explanations38.
For children with ADHD, AI-driven digital therapy and gamification platforms like Dysolve AI transform boring exercises into highly engaging tasks. Meta-analyses have shown that gamification interventions that require rapid reaction, planning, or sustained attention can significantly improve working memory and emotional stability in these children 41 . In terms of autism (ASD) intervention, systems based on AI emotion recognition technology can help children with ASD practice emotional regulation and social skills in a controlled environment 1. A multiple-probe across participants single-case design in a preschool inclusive classroom proved that the AI-supported self-coaching system can significantly improve the fidelity of teachers' implementation of embedded instruction (Embedded Instruction) for children with ASD, thereby significantly improving children's learning outcomes and target response rates 42.
Effect size differences and limitations of empirical research
Data based on a systematic literature review confirm the quantitative advantages of AI in special education. Empirical research shows that AI education systems produce a huge effect size (Effect Size, measured by Cohen's ![][image1]) in improving the skills of students with learning disabilities. In the evaluation of an AI system for mathematical arithmetic exercises, the intervention effect size for the general student group was ![][image2] (already a significantly large effect), while the effect size for students with mathematics learning disabilities was as high as ![][image3] 8. This data eloquently proves that AI adaptive systems can not only bridge the ability gap, but also achieve unexpected intervention effects in specific cognitive areas.
In addition, AI technology solves the high cost and difficulty of scaling up problems of traditional one-to-one tutoring in the "high-dose tutoring" (high-dose tutoring) model, which is intensive group tutoring at least three times a week for 30 minutes each time 9. The Intelligent Tutoring System (ITS) uses natural language processing and real-time student modeling to dynamically adjust the difficulty of questions to keep them within Vygotsky's "Zone of Proximal Development" (ZPD), ensuring the timeliness and pertinence of intervention 1.
However, risk of bias (Risk of Bias) assessment of current academic literature reveals blind spots worthy of vigilance. A systematic review using ROBINS-I and JBI critical assessment tools showed that among existing empirical studies on AI in special education, up to 70% of studies had moderate risk of bias, 30% had high/severe risk of bias, and no study was rated as low risk 8 . Many studies lack long-term longitudinal follow-up (Longitudinal Follow-up), making it difficult for academics to judge whether short-term skill improvement will bring long-term negative effects of "cognitive offloading"8.
Algorithmic Bias, Data Fairness and Inequality in Language Technology
While AI promotes the personalization of education, it may also become an amplifier of structural inequality. Algorithmic systems are not created in a vacuum; their underlying training data, model architecture, and evaluation criteria often contain historically accumulated social biases 44 .
Systematic Oppression of Children’s Speech and Dialect Recognition
Automatic speech recognition (ASR) technology is widely used in language learning and early reading assessment. However, because most commercial ASR systems rely too much on standard accents and adult speech data during training, they show significant accuracy decline when faced with children's complex and varied acoustic features (such as higher frequency spectrum, varied prosody, irregular and complex syntax) 47 .
Dialect and racial differences further exacerbate this technological divide. A sociolinguistic evaluation of the Newcastle English corpus in the UK Tyneside Historical Electronic Corpus (DECTE) showed that top commercial ASR systems produced more than 3,000 serious errors in transcription 49 . These errors do not occur randomly, but show a clear social class distribution - with a high concentration on specific vowel quality (Vowel Quality) and glottalization features (Glottalisation), with error rates being highest at both ends of the age spectrum (children and older adults) 49 . Similarly, research shows that the recognition accuracy of ASR systems for African-American English and black tutors is significantly lower than that of white standard English 50 .
In educational applications, this means that groups of children who speak dialects, are not native speakers of English, or whose phonology is immature will systematically receive lower scores or poorer feedback in automated speech-based reading assessments or diagnostic systems for language disorders (e.g., SSD, Speech Sound Disorder) 44 . This phenomenon constitutes “predictive bias”, which is essentially the systematic oppression of disadvantaged groups at the technical level53. To alleviate this problem, organizations such as SoapBox Labs have begun building deep learning models specifically based on thousands of hours of children's real speech (reflecting diverse populations, accents, and speaking styles) to reduce the risk of AI bias in educational evaluation 40 .
The introduction of critical AI literacy and algorithmic auditing
Given the inevitability of generative AI bias (stemming from data, English hegemony, and capitalist logic) 46 , there is an urgent need for the education community to integrate “Algorithm Auditing” and Critical AI Literacy into K-12 curricula 45 . This requires teachers to guide students not only to learn how to use AI, but also to explore the data sources, fairness principles and accountability mechanisms behind AI decision-making 55 . By engaging students in activities such as testing simulated loan application algorithms and taking gallery walks to analyse bias cases, they can move digital citizenship from theory to practice and learn to question rather than blindly accept the output of automated systems. 44
Comparison and integration of human teacher role transformation and AI feedback mechanism
In education scenarios, the intervention of AI has fundamentally changed the feedback ecosystem. Regarding the mechanism of AI feedback and human teacher feedback, the academic and practical fields are highly complementary.
| Assessment dimensions | AI-generated feedback (GenAI Feedback) | Human teacher feedback (Human Teacher Feedback) | Source basis |
|---|---|---|---|
| Speed, Scale & Access | Extreme: Process large-scale student work instantly and provide massive amounts of detail, 24/7. | Lower: limited by class size, teaching load and personal energy. | 57 |
| Objectivity and Risk Perception | High: Some students believe that they are not affected by interpersonal biases and there is a low psychological risk in asking for help. | Medium/High: Relies on the establishment and maintenance of trusting relationships between teachers and students. | 60 |
| Emotional Support and Empathy | Very Low: Lack of emotional nuance and context-specific resonance, appearing cold. | Extremely high: Provides irreplaceable emotional connection, security and psychological counseling. | 58 |
| Higher-order thinking and motivation stimulation | Limited: Good at correcting basic grammar and fact-checking, but not good at in-depth enlightenment. | Core: Good at stimulating deep motivations through explanation, questioning and personal charm. | 57 |
An empirical study of Makerspaces conducted by the Harvard Graduate School of Education provided rare causal evidence: by using the large language model (GPT-3) as a "backstage coach" for teachers, teachers' uptake of student ideas increased by about 10%, which was mainly attributed to teachers raising higher quality follow-up questions 62. This shows that the most effective application of AI is not to directly replace teachers with students, but to empower teachers.
The OECD divides the cooperation models between teachers and AI into three types: replacement, complementarity and augmentation. Research has unanimously pointed out that the "enhanced mode" is the most effective path. It can save about 31% of teachers' administrative and regular correction time, allowing teachers to reinvest their energy in instructional design, innovation, and establishing deep emotional connections with students 13.
In German educational practice, this integration model is being verified regionally. For example, the AI feedback tool “fiete.ai” was piloted extensively in the KIMADU project in Saxony-Anhalt and North Rhine-Westphalia (NRW) 59 . The tool connects to the GPT-4 language model, allowing students to scan compositions and receive immediate formative feedback based on specific criteria (e.g., logic, grammar). Teachers can view students’ modification tracks and ability shortcomings through digital dashboards (such as progress bars of different colors) 59 . However, the education union (GEW/VBE) keenly pointed out that the application of this tool does not directly reduce teachers’ workload, because teachers must spend extra time checking AI “hallucinations” (Hallucinations) and providing psychological support, proving that “human-machine collaboration” is not only a technical implementation, but also a complex reconstruction of the educational ecology 59 .
Regulatory Framework, Data Privacy and Implementation of the United Nations Principles on the Rights of the Child
Faced with the capital expansion and technological black box of multinational technology companies, how to build regulatory guardrails (Guardrails) to protect children's rights and interests has become the focus of global policy games. Since children do not yet have the cognitive ability to fully understand "privacy" in an abstract manner (for example, children under 11 years old often deliberately click on links with age restriction warnings out of curiosity), relying entirely on user self-protection has failed 6.
The Digital Extension of the United Nations Framework and the Convention on the Rights of the Child
In 2026, the United Nations Committee on the Rights of the Child (CRC), the International Telecommunication Union (ITU), the United Nations Children's Fund (UNICEF) and other institutions issued the landmark "Joint Statement on Artificial Intelligence and the Rights of the Child" 12. This statement builds on the United Nations Convention on the Rights of the Child (UNCRC) and its General Comment No. 25 on the Digital Environment, establishing the world’s first international framework dedicated to children’s AI rights12.
The Joint Declaration clearly states that children’s rights not only apply in the physical world, but also have full validity in the AI-driven digital environment5. The statement established 11 priority areas for action, emphasizing that AI must be designed and deployed with "Best interests of the child" as the core principle. It sends an extremely strong political signal: commercial profits must not be at the expense of children's rights. It requires governments and technology companies to establish a global defence line that includes accountability, transparency, privacy protection, non-discrimination and freedom from exploitation, and clearly defines the negative and positive obligations of countries and companies. 5
Conflict and reconciliation between the EU AI Act and German data privacy regulations in schools
Globally, the EU's regulatory framework is at the forefront. The newly effective European Artificial Intelligence Act (EU AI Act) adopts a risk-based governance model (Risk-Based Approach) 72. The bill comprehensively bans unacceptable threats (such as technologies that exploit the psychological vulnerability of children for subconscious manipulation), mandates transparency, and strictly regulates the application of high-risk AI in educational scenarios (such as automated decision-making systems for student assessment and educational diversion), while requiring the implementation of a strict age verification framework 72.
When it comes to data privacy, the General Data Protection Regulation (GDPR) sets high standards for children’s data protection. In Germany, its Federal Data Protection Act (BDSG) strengthens the local protection net through specific exemptions and supplementary provisions based on the full implementation of the GDPR. For example, Article 42 of the BDSG clearly stipulates criminal liability (Criminal Penalties) for illegal commercial transmission of large-scale personal data, and the threshold for appointing a Data Protection Officer (DPO) is more stringent76. However, at the actual school operation level, extremely strict privacy protection often collides violently with the innovation needs of digital teaching, forming the so-called “data protection versus functionality dilemma”79.
In order to solve this dilemma, the German Data Protection Conference (DSK) proposed an innovative solution, recommending to amend the relevant provisions of the GDPR to transfer the legal compliance responsibility for "Privacy by Design and by Default" from schools as users to "manufacturers and providers" of standard IT products (just like the "Cyber Resilience Act" CRA), thus significantly reducing the compliance pressure on schools 80. At the same time, the Joint Meeting of German State Ministers of Education (KMK) issued comprehensive guidelines for the use of AI in schools between 2024 and 2025, and states such as North Rhine-Westphalia (NRW) and Bremen have also issued specific operational rules 56 .
In these regional practices, policies require not only the use of AI (Learning with AI), but also the use of AI as a teaching object (Learning about AI). The DPACK (Digitality-Related Pedagogical and Content Knowledge) model evolved on the basis of the TPACK theoretical framework, as well as the AI competency framework of the United Nations Educational, Scientific and Cultural Organization (UNESCO), both emphasize that AI ethics, bias identification and technical operation principles should be popularized as basic literacy (AI Literacy) in primary and basic education stages to ensure that children become shapers of the digital world rather than passive consumers 56 . The Education Guidelines emphasize that the use of any AI tools (such as automatic translation, grammatical reconstruction, feedback generation) must meet strict citation transparency and cannot replace traditional education standards and the "unassisted examination" system 56.
Timetable and future prospects of educational technology evolution
Looking at the timeline of technological development, we are on the eve of a profound change. According to industry forecasts, between 2025 and 2027, multimodal AI (Multimodal AI, integrating text, voice, images and virtual space) and intelligent adaptive learning platforms will become the mainstream of K-12 education, and AI tutors will provide context-aware support closer to human levels 2. From 2027 to 2030, the in-depth combination of virtual reality (VR) and AI, cognitive status monitoring with biometric tracking, and more secure federated learning technology will achieve the ultimate personalized education while solving data privacy problems2. Brain-computer interfaces (BCIs) and specialized educational AI chips are also regarded as frontier areas of exploration with high potential2.
In this new era of human-machine symbiosis, the core proposition is not how to infinitely increase the intelligence limit of machines, but how to defend the "human dimension" in education. Educational policymakers, technology developers, and frontline educators must establish an interdisciplinary collaboration mechanism to effectively transform the macro-regulatory spirit of the United Nations’ Joint Statement on Artificial Intelligence and Children’s Rights and the EU AI Act into the micro-ethical design of every educational software code. The future education system should resolutely implement the "augmentation" human-computer collaboration model, hand over the efficiency of imparting conventional knowledge and correcting grammatical fallacies to sleepless algorithms, and return empathic understanding, ethical guidance, value recognition and the awakening of critical souls to human teachers. Only under the premise of ensuring that education is fair to every marginalized group, that technological transparency can be strictly audited by the public, and that children's privacy and free development rights are absolutely protected, can artificial intelligence truly transcend the "learning-performance paradox" and become an empowering light that lights up children's future, rather than a computing black box that deprives them of independent thinking and social growth experience.
Works Cited
1. The Use of Artificial Intelligence (AI) in Early Childhood Education, accessed May 12, 2026, https://www.mdpi.com/2075-4698/15/12/341
2. Strategic Integration of Artificial Intelligence in U.S. K–12 Education \- Preprints.org, accessed May 12, 2026, https://www.preprints.org/frontend/manuscript/9482bd844dac55a121291e236882fb98/download\_pub
3. Artificial Intelligence and the Future of Teaching and Learning (PDF) \- Department of Education, accessed May 12, 2026, https://www.ed.gov/sites/ed/files/documents/ai-report/ai-report.pdf
4. “Screen Time” to “AI Time”: AI Use and Cognitive, Emotional, and Behavioral Outcomes in Children and Adolescents \- PMC, accessed May 12, 2026, [https://pmc.ncbi.nlm.nih.gov/articles/PMC12893840/](https://pmc.
ncbi.nlm.nih.gov/articles/PMC12893840/)
5. Joint statement on artificial intelligence and the rights of the child | Together Scotland, accessed May 12, 2026, [https://www.togetherscotland.org.uk/news-and-events/news/2026/2/joint-statement-on-artificial-intelligence-and-the-rights-of-the-child/] (https://www.togetherscotland.org.uk/news-and-events/news/2026/2/joint-statement-on-artificial-intelligence-and-the-rights-of-the-child/)
6. The Future of Child Development in the AI Era: \- | Everyone.AI, accessed May 12, 2026, https://everyone.ai/wp-content/uploads/2024/05/EveryoneAI.ResearchPaper.pdf
7. Application of artificial intelligence in early childhood development: a scoping review protocol \- PMC, accessed May 12, 2026, https://pmc.ncbi.nlm.nih.gov/articles/PMC12587935/
8. The Effectiveness of Artificial Intelligence
lligence-Based Interventions for Students with Learning Disabilities: A Systematic Review \- PMC, accessed May 12, 2026, https://pmc.ncbi.nlm.nih.gov/articles/PMC12385150/
9. The Transformative Power of AI-Enhanced High-Dose Tutoring | NORC at the University of Chicago, accessed May 12, 2026, https://www.norc.org/research/library/unlocking-hearts-and-minds-transformative-power-of-ai-enhanced-high-dose-tutoring.html
10. CDT Survey Research Finds Use of AI in K-12 Schools Connected to Negative Effects on Students, Including Their Real-Life Relationships \- Centre for Democracy and Technology, accessed May 12, 2026, [https://cdt.org/press/cdt-survey-research-finds-use-of-ai-in-k-12-schools-connected-to-negative-effects-on-students-including-their-real-life-relationship
s/](https://cdt.org/press/cdt-survey-research-finds-use-of-ai-in-k-12-schools-connected-to-negative-effects-on-students-including-their-real-life-relationships/)
11. Artificial Intelligence in Classrooms: Cognitive Dimensions \- European Parliament, accessed May 12, 2026, [https://www.europarl.europa.eu/RegData/etudes/BRIE/2026/784575/IUST\_BRI(2026)784575\_EN.pdf] (https://www.europarl.europa.eu/RegData/etudes/BRIE/2026/784575/IUST_BRI\(2026\)784575_EN.pdf)
12. Child rights and AI: CRC adopts first global framework with partners | OHCHR, accessed May 12, 2026, https://www.ohchr.org/en/statements/2026/02/child-rights-and-ai-crc-adopts-first-global-framework-partners
13. 355\. What the OECD's New 2026 AI Report Means for Every Teacher (Key Findings), accessed May 12, 2026, [https://www.youtube.com/watch?v=jzeUxRBpVZI](https://www.youtub
e.com/watch?v=jzeUxRBpVZI)
14. OECD Digital Education Outlook 2026, accessed May 12, 2026, https://www.oecd.org/en/publications/oecd-digital-education-outlook-2026\_062a7394-en.html
15. OECD Digital Education Outlook 2026: how generative AI can support learning when used with purpose, accessed May 12, 2026, [https://digital-skills-jobs.europa.eu/en/latest/news/oecd-digital-education-outlook-2026-how-generative-ai-can-support-learning-when-used] (https://digital-skills-jobs.europa.eu/en/latest/news/oecd-digital-education-outlook-2026-how-generative-ai-can-support-learning-when-used)
16. How AI quietly undermines the joy and effort of learning: a call for rebalancing education in the digital age \- PMC, accessed May 12, 2026, https://pmc.ncbi.nlm.nih.gov/articles/PMC12333830/
17. Children's Overtrus
t: Intentional Use of Robot Errors to Decrease Trust \- University of Manitoba, accessed May 12, 2026, http://hci.cs.umanitoba.ca/assets/publication\_files/Childrens\_Overtrust.pdf
18. A race to belief: How Evidence Accumulation shapes trust in AI and Human informants, accessed May 12, 2026, https://arxiv.org/html/2511.22617v1
19. Artificial Intelligence and Youth: Cognitive, Educational, and Behavioral Impacts \- MDPI, accessed May 12, 2026, https://www.mdpi.com/2673-2688/7/4/121
20. Is AI dulling our minds? - Harvard Gazette, accessed May 12, 2026
21. Best Smart AI Robots for Kids (2025 Picks) \- JetLearn, accessed May 12, 2026, [https://www.j
etlearn.com/blog/7-smart-ai-robots-for-kids-that-teach-entertain](https://www.jetlearn.com/blog/7-smart-ai-robots-for-kids-that-teach-entertain)
22. Parents Raising Successful Kids Swear By These Smart Robot Toys for Teaching AI, accessed May 12, 2026, https://www.aifunlab.io/learn/smart-robot-kids-for-kids
23. Miko AI-Powered Robot: Smart Learning Partner for Kids, accessed May 12, 2026, https://miko.ai/
24. AI Robot Toy: Best Picks for Smart Kids in 2026 | Loona Blog, accessed May 12, 2026, https://keyirobot.com/blogs/buying-guide/ai-robot-toy-best-picks-for-smart-kids-in-2025
25. Children's attribution of mental states to humans and social robots assessed with the Theory of Mind Scale \- PMC, accessed May 12, 2026, [https://pmc.ncbi.nlm.nih.gov/articles/PMC12065903/](https:/
/pmc.ncbi.nlm.nih.gov/articles/PMC12065903/)
26. Child–Robot Relationship Formation: A Narrative Review of Empirical Research \- PMC, accessed May 12, 2026, https://pmc.ncbi.nlm.nih.gov/articles/PMC7235061/
27. Preschoolers' anthropomorphizing of robots: Do human-like properties matter? \- Frontiers, accessed May 12, 2026, https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2022.1102370/full
28. Anthropomorphism in human–robot interactions: a multidimensional conceptualization | Communication Theory | Oxford Academic, accessed May 12, 2026, https://academic.oup.com/ct/article/33/1/42/6762954
29. Kids' Trust in Robots vs Humans: A Study in Selective Learning \- Neuroscience News, accessed May 12, 2026, [https://neurosciencenews.co
m/robot-learning-child-development-25394/](https://neurosciencenews.com/robot-learning-child-development-25394/)
30. Social Robots and Children: A Field in Development \- UvA-DARE (Digital Academic Repository), accessed May 12, 2026, https://pure.uva.nl/ws/files/230055708/Social\_Robots\_and\_Children.pdf
31. The robot that stayed: understanding how children and families engage with a retired social robot \- PMC, accessed May 12, 2026, https://pmc.ncbi.nlm.nih.gov/articles/PMC12370748/
32. Child-Robot Interaction | American Scientist, accessed May 12, 2026, https://www.americanscientist.org/article/child-robot-interaction
33. The effects of screen time on children: The latest research parents should know - CHOC, accessed May 12, 2026, [https://health.
choc.org/the-effects-of-screen-time-on-children-the-latest-research-parents-should-know/](https://health.choc.org/the-effects-of-screen-time-on-children-the-latest-research-parents-should-know/)
34. The impact of digital activities on children's lives: How's Life for Children in the Digital Age? | OECD, accessed May 12, 2026, https://www.oecd.org/en/publications/how-s-life-for-children-in-the-digital-age\_0854b900-en/full-report/the-impact-of-digital-activities-on-children-s-lives\_4df70664.html
35. The Impact of AI on Children's Development | Harvard Graduate School of Education, accessed May 12, 2026, https://www.gse.harvard.edu/ideas/edcast/24/10/impact-ai-childrens-development
36.Ex
ploring the Role of Artificial Intelligence in Shaping Students' Social Skills \- ERIC, accessed May 12, 2026, https://files.eric.ed.gov/fulltext/ED679024.pdf
37. AI and Early Childhood Education: A Need for Robust Safeguards, accessed May 12, 2026, https://www.instituteforchildsuccess.org/ai-early-childhood-education/
38. AI in the Classroom: Creating New Opportunities for Students with Special Needs, accessed May 12, 2026, https://www.specialneedsalliance.org/blog/ai-in-the-classroom-creating-new-opportunities-for-students-with-special-needs/
39. Best AI Tools for Dyslexia in 2025 | AIapps, accessed May 12, 2026, [https://www.aiapps.com/blog/ai-tools-dyslexia/](https://www.aiapps.com/blog/ai-tools-d
yslexia/)
40. AI in Education: Examples from the Field, Including Voice Technology \- ReadSpeaker, accessed May 12, 2026, https://www.readspeaker.com/blog/voice-technology-in-education/
41. AI for the Neurodiverse Classroom: How AI Tools Help Students With ADHD and Dyslexia, accessed May 12, 2026, https://edcircuit.com/ai-for-the-neurodiverse-classroom-how-ai-tools-help-students-with-adhd-and-dyslexia/
42. Investigating the Impact of AI-Supported Self-Coaching as a Professional Development Model for Embedded Instruction in Inclusive Early Childhood Settings \- PMC, accessed May 12, 2026, https://pmc.ncbi.nlm.nih.gov/articles/PMC12837420/
43. A Comprehensive Review of AI-based Intelligent Tutoring Systems: Applicat
ions and Challenges \- arXiv, accessed May 12, 2026, https://arxiv.org/html/2507.18882v1
44. Teaching students about algorithmic bias through real-world examples \- SchoolAI, accessed May 12, 2026, https://schoolai.com/blog/algorithmic-bias-examples-education
45. Exploring Strategies to Detect and Mitigate Bias in AI in Education: Students' Perceptions and Didactic Approaches \- MDPI, accessed May 12, 2026, https://www.mdpi.com/2813-4346/5/2/33
46. Helping students understand the biases in generative AI \- Centre for Teaching Excellence, accessed May 12, 2026, https://cte.ku.edu/addressing-bias-ai
47. Closing the Child Speech Recognition Gap: Evidence, Limitations, and Paths Forward, accessed May 12, 2026, [https://the-learning-agency.com/guides-resources/closing-th
e-child-speech-recognition-gap-evidence-limitations-and-paths-forward/](https://the-learning-agency.com/guides-resources/closing-the-child-speech-recognition-gap-evidence-limitations-and-paths-forward/)
48. What Happens When AI Doesn't Understand Students? An example for creative and equitable AI policy in education | Getting Smart, accessed May 12, 2026, [https://www.gettingsmart.com/2023/01/31/what-happens-when-ai-doesnt-understand-students-an-example-for-creative-and-equitable-ai-policy-in-education/] (https://www.gettingsmart.com/2023/01/31/what-happens-when-ai-doesnt-understand-students-an-example-for-creative-and-equitable-ai-policy-in-education/)
49. A Sociolinguistic Analysis of Automatic Speech Recognition Bias in Newcastle English, accessed May 12, 2026, https://arxiv.org/html/2603.24549v1
50. Tackling Bias in Automatic Speech Recognition \- Two Examples From Our Ongoing Work, accessed May 12, 2
026, [https://www.colorado.edu/research/ai-institute/2025/01/22/tackling-bias-automatic-speech-recognition-two-examples-our-ongoing-work] (https://www.colorado.edu/research/ai-institute/2025/01/22/tackling-bias-automatic-speech-recognition-two-examples-our-ongoing-work)
51. Toward Responsible ASR for African American English Speakers: A Scoping Review of Bias and Equity in Speech Technology \- AAAI Publications, accessed May 12, 2026, https://ojs.aaai.org/index.php/AIES/article/download/36580/38718/40655
52. Artificial intelligence (AI)-driven technologies for managing pediatric speech and language therapy: A scoping review \- PMC, accessed May 12, 2026, https://pmc.ncbi.nlm.nih.gov/articles/PMC12589799/
53. understanding algorithmic bias in British English automatic speech recognition \- ACM FAccT, accessed May 12, 2
026, https://facctconference.org/static/pdfs\_2022/facct22-3533117.pdf
54. Health advisory: Artificial intelligence and adolescent well-being \- American Psychological Association, accessed May 12, 2026, https://www.apa.org/topics/artificial-intelligence-machine-learning/health-advisory-ai-adolescent-well-being
55. Empowering Teens to Defang Bias in AI with Algorithm Auditing, accessed May 12, 2026, https://csteachers.org/empowering-teens-to-defang-bias-in-ai-with-algorithm-auditing/
56. Empfehlung zum Umgang mit auf künstlicher Intelligenz ..., accessed May 12, 2026, [https://www.bildung.bremen.de/sixcms/media.php/13/20250602\_KI-Handreichung\_final.pdf](https://www.bildung.bre
men.de/sixcms/media.php/13/20250602_KI-Handreichung_final.pdf)
57. How Does AI Driven Feedback Differ from Traditional Teacher Feedback? \- Apporto, accessed May 12, 2026, https://www.apporto.com/ai-driven-feedback-vs-traditional-teacher-feedback
58. Human and AI-generated feedback in higher education: A systematic review of effectiveness and student perceptions, accessed May 12, 2026, https://www.cedtech.net/download/human-and-ai-generated-feedback-in-higher-education-a-systematic-review-of-effectiveness-and-student-17863.pdf
59. KI-Tool löst Diskussion über Schule aus \- Deutschdidaktik, accessed May 12, 2026, [https://deutschdidaktik.germanistik.uni-halle.de/ki-tool-loest-diskussion-ueber-schule-aus/](https://deutschdi
daktik.germanistik.uni-halle.de/ki-tool-loest-diskussion-ueber-schule-aus/)
60. Full article: Comparing Generative AI and teacher feedback: student perceptions of usefulness and trustworthiness \- Taylor & Francis, accessed May 12, 2026, https://www.tandfonline.com/doi/full/10.1080/02602938.2025.2502582
61. Impact Of Artificial Intelligence On The Student–Teacher Relationship, accessed May 12, 2026, https://ijrtssh.com/2026/05/impact-of-artificial-intelligence-on-the-student-teacher-relationship/
62. Can AI Feedback Support More Responsive Teaching? \- The Chronicle of Evidence-Based Mentoring, accessed May 12, 2026, [https://www.evidencebasedmentoring.org/helping-teachers-hear-students-with-help-from-ai/](https://www.evidencebasedmentoring.org/helping-teachers-hear-students
s-with-help-from-ai/)
63. Using Artificial Intelligence to Optimize Personalized Learning \- KnowledgeWorks, accessed May 12, 2026, https://knowledgeworks.org/resources/artificial-intelligence-optimize-personalized-learning/
64. Für einen kritisch-konstruktiven Umgang mit KI in der Schule – Neue ..., accessed May 12, 2026, https://www.schulministerium.nrw/presse/pressemitteilungen/fuer-einen-kritisch-konstruktiven-umgang-mit-ki-der-schule-neue
65. Fiete.ai in der Deutsch-Schulaufgabe \- IQES, accessed May 12, 2026, https://www.iqesonline.net/blogs/fiete-ai-in-der-deutsch-schulaufgabe/
66. E 18/1416 \- Landtag NRW, accessed May 12, 2026, [https://www.land
tag.nrw.de/portal/WWW/dokumentenarchiv/Dokument/esm/MME18-1416.pdf](https://www.landtag.nrw.de/portal/WWW/dokumentenarchiv/Dokument/esm/MME18-1416.pdf)
67. Joint Statement on Artificial Intelligence and the Rights of the Child \- Knowledge @ UNICEF, accessed May 12, 2026, https://knowledge.unicef.org/resource/joint-statement-artificial-intelligence-and-rights-child
68. Joint Statement on Artificial Intelligence and the Rights of the Child, accessed May 12, 2026, https://www.asamblea.go.cr/sd/referencia\_cedil/2026/Alerta\_feb\_26/110226.pdf
69. General comment No. 25 (2021) on children's rights in relation to the digital environment, accessed May 12, 2026, [https://www.ohchr.org/en/documents/general-comments-and-recommendations/general-comment-no-25-2021-childrens-rig
hts-relation](https://www.ohchr.org/en/documents/general-comments-and-recommendations/general-comment-no-25-2021-childrens-rights-relation)
70. Children are not test subjects: Joint Statement reaffirms children's rights in the AI era, accessed May 12, 2026, https://5rightsfoundation.com/children-are-not-test-subjects-joint-statement-reaffirms-childrens-rights-rights-in-the-ai-era/
71. General Comment No. 25 on Children's Rights in Relation to the Digital Environment: Implications for Children's Right to Privacy and Data Protection in Africa | Human Rights Law Review | Oxford Academic, accessed May 12, 2026, https://academic.oup.com/hrlr/article/24/3/ngae018/7692952
72. AI's future for students is in our hands | Brookings, accessed May 12, 2026, [https://www.brookings.e
du/articles/ais-future-for-students-is-in-our-hands/](https://www.brookings.edu/articles/ais-future-for-students-is-in-our-hands/)
73. Mind the Gap, accessed May 12, 2026, https://www.interface-eu.org/publications/age-assurance-gap
74. Protecting children online: Selected EU, national and regional laws and initiatives \- European Parliament, accessed May 12, 2026, [https://www.europarl.europa.eu/RegData/etudes/BRIE/2025/769570/EPRS\_BRI(2025)769570\_EN.pdf] (https://www.europarl.europa.eu/RegData/etudes/BRIE/2025/769570/EPRS_BRI\(2025\)769570_EN.pdf)
75. Grundschule aktuell, accessed May 12, 2026, https://grundschulverband.de/wp-content/uploads/2025/05/GSa170\_Mai25\_ES-1.pdf
76. Data protection laws in Germany, accessed May 12, 2026, [https://www.dlapiperdataprotection.com/?c=DE](https://www.dlapiperdat
aprotection.com/?c=DE)
77. German Federal Data Protection Act (BDSG): Impact, Benefits & Compliance \- Kiteworks, accessed May 12, 2026, https://www.kiteworks.com/risk-compliance-glossary/german-federal-data-protection-act-bdsg/
78. Children's Online Privacy and Data Protection in Selected European Countries \- Loc, accessed May 12, 2026, https://tile.loc.gov/storage-services/service/ll/llglrd/2021680641/2021680641.pdf
79. Data protection versus functionality: the dilemma of German schools | Heinrich Böll Stiftung, accessed May 12, 2026, https://www.boell.de/en/2021/04/15/Data-protection-versus-functionality-the-dilemma-of-German-schools
80. German Government P
roposes GDPR Reform to Shift Responsibility to Manufacturers, accessed May 12, 2026 https://www.insideprivacy.com/european-union-2/german-government-proposes-gdpr-reform-to-shift-responsibility-to-manufacturers/
81. Künstliche Intelligenz in der Grundschule \- DGfE, accessed May 12, 2026, https://www.dgfe.de/fileadmin/OrdnerRedakteure/Sektionen/Sek05\_SchPaed/GFPP/Positionspapier\_KI-in-der-GS.pdf
82. Generative AI in K-12 Education: Transforming Learning Through Innovation \- MAKEBOT.AI, accessed May 12, 2026, [https://www.makebot.ai/blog-en/generative-ai-in-k-12-education-transforming-learning-through-innovation](https://www.makebot.ai/blog-en/generative-ai-in-k-12-education-transforming-learning-throug
h-innovation)
[image1]: <data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAABIAAAAUCAYAAACAl21KAAAAbElEQVR4XmNgGAWDCmgC8X8ophh4AvEDdEFywHMgdkEXJBYcB+I tQHyFAeItDlRp4oAZEGtD2SCX/EOSIwkgB2wrEG9F4hMNJBlQDSI7fNANgoXPaSQxokEeEL8CYnsGSGC/B2JWFBWjYBgAAH58FDqv+qafAAAAAElFTkSuQmCC>
[image2]: <data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAE4AAAAaCAYAAAAZtWr8AAACZ0lEQVR4Xu2XTchNQRjHH6F8bVC8opStBeUrZektEgsU CzsLkiSKspWNlSws9JasLFhSksWJpY0FKTs2FhaysEA+/r87M85zxzn3vb33nJfF/OrfnXmeOTNznjvzzByzQqFQKBQK/4bV0jrpqLQr8xVGsEY6KH2 RBep7ZWO3rM5tnm3Qhs1VRsFv6Km1ITrHJwmrm2U5ZK223sJx5GT9oX6ywMNbdzM4L5oHxEASem7Z6Zf6Qjscy/eGn/8RK6YU042wTcUB6Jk1ZWMYMi EbBhDhAxtGoLYe/KXDkrNyWc9/quZLLFjpfFe0+cJSxo04gv2x2dfIAGsVF6f2YIh+1MUngllkduHyHVNHWW+BY2uQTDwNyMMwHcw0cQUgHCivtm4V+ XkVbFeu9BI4TkwC9zmwMOF8HA3m1LXA3M5vntHTC1VltaeUBh1tb4CZeFPzbTNAn4b0WEjPJd5W02Pk8XW1V4AWrzEbOYi5NpHn7oADtU4rhneiXton 03ClnmxPpH+Cym6Bj7nDwQFrkfH0x2z2OX+qcmql+QzoW64mr0sdYbrrHHbLQTyf3OJb7nVjeYSFwbFOCejg16pkjFoKSXmiL9LZ2D+6VfhvCRgvBTs FNee78nxZmO234fvjS/v6DJoK8xmdWOs7JO7Ntrz7gs+i6jf99zHwvWfjk8rvGQ9BvWciXbWmnUCgUCoVC4X/lN6fklnY+7NkXAAAAAAElFTkSuQmCC>
[image3]: <data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAE4AAAAaCAYAAAAZtWr8AAACPUlEQVR4Xu2XP0gcQRTGn0TBoPgv4h8QBEmXQ kRNECxSBEEkGCSCQroUaYKFFoK9lY2IKNiEtCFtGkkhlrESFGslIBhCKlMoiX5f5k3u3eLdzXm3e8384IN582Z3Z9/Mm7crEolEIpFIpDY8g nqh19CzhC9ShC7oJXQNjSd8aTIN3Yh7dign0G9oBfoBPTW+BugKegftaHvB+FNhFfoCNSYdKfAceiXupcsJHAP1WNvMEF57qHYndAwdqU2+i xvTbvqqShP0VdwqZsmwhAeuXtxYT7Pa22ozkKfQ5f8RzuYY+qpKNzQCTYp7QF++O3XKCdwy9FPbTMk64ysE722DXTFT0D7UAz2UsAdwhblyI Qp5KRIaOKbaAXQBvYUeQPPisqTVjCM8auagb9AnqCPfXRm/oCfG/qsqxhJ0FihW6RBCA+fTkGOZsoSLY1PV80LcHFgYuEEY5KrA1WD1tHACL AxZU27gzhP9pTKFu47+0IUsCLcxA8TqY/t486wLAwkNHI+JPXHBs5QK3IY4//uko1z8yi2aPm5truSAuPOAB+9d1DJVCYtDscCta5vz9PAa9 vEsrAi/cnainAy/4chnyZ0hWVAocP78+mP6fBGz0N7V9qbaMzn3v3dl35jpuzdvoA/aHhUXOKYpg2ofmiZt4j591sS92EfJr8b8rrwrDVnU+ rU9qLb/cxhSu0Vt3osFj/cIrfIl4bnG3yxfcfg9F5petWYW2oImkg6Fu5i/WzzfsvgLikQikUgkEqkmt5oUikK16LMYAAAAAElFTkSuQmCC>